Skip to main content

harn_cli/
lib.rs

1#![recursion_limit = "256"]
2
3pub mod acp;
4pub mod cli;
5pub mod commands;
6pub mod config;
7pub mod env_guard;
8pub mod format;
9pub mod package;
10mod provider_bootstrap;
11pub mod skill_loader;
12pub mod skill_provenance;
13pub mod test_runner;
14#[doc(hidden)]
15pub mod tests;
16
17use clap::{error::ErrorKind, CommandFactory, Parser as ClapParser};
18use std::path::{Path, PathBuf};
19use std::sync::Arc;
20use std::{env, fs, process, thread};
21
22use cli::{
23    Cli, Command, CompletionShell, MergeCaptainCommand, MergeCaptainMockCommand, ModelInfoArgs,
24    PackageArtifactsCommand, PackageCacheCommand, PackageCommand, PersonaCommand,
25    PersonaSupervisionCommand, ProvidersCommand, RunsCommand, ServeCommand, SkillCommand,
26    SkillKeyCommand, SkillTrustCommand, SkillsCommand, ToolCommand,
27};
28use harn_lexer::Lexer;
29use harn_parser::{DiagnosticSeverity, Parser, TypeChecker};
30
31pub const CLI_RUNTIME_STACK_SIZE: usize = 16 * 1024 * 1024;
32
33#[cfg(feature = "hostlib")]
34pub(crate) fn install_default_hostlib(vm: &mut harn_vm::Vm) {
35    let _ = harn_hostlib::install_default(vm);
36}
37
38#[cfg(not(feature = "hostlib"))]
39pub(crate) fn install_default_hostlib(_vm: &mut harn_vm::Vm) {}
40
41/// Entry point used by `src/main.rs`. Hosts the CLI runtime thread and
42/// drives the async dispatcher in `async_main`.
43pub fn run() {
44    let handle = thread::Builder::new()
45        .name("harn-cli".to_string())
46        .stack_size(CLI_RUNTIME_STACK_SIZE)
47        .spawn(|| {
48            let runtime = tokio::runtime::Builder::new_multi_thread()
49                .enable_all()
50                .build()
51                .unwrap_or_else(|error| {
52                    eprintln!("failed to start async runtime: {error}");
53                    process::exit(1);
54                });
55            runtime.block_on(async_main());
56        })
57        .unwrap_or_else(|error| {
58            eprintln!("failed to start CLI runtime thread: {error}");
59            process::exit(1);
60        });
61
62    if let Err(payload) = handle.join() {
63        std::panic::resume_unwind(payload);
64    }
65}
66
67async fn async_main() {
68    let raw_args = normalize_serve_args(env::args().collect());
69    if raw_args.len() == 2 && raw_args[1].ends_with(".harn") {
70        provider_bootstrap::maybe_seed_ollama_for_run_file(Path::new(&raw_args[1]), false, false)
71            .await;
72        commands::run::run_file(
73            &raw_args[1],
74            false,
75            std::collections::HashSet::new(),
76            Vec::new(),
77            commands::run::CliLlmMockMode::Off,
78            None,
79            commands::run::RunProfileOptions::default(),
80        )
81        .await;
82        return;
83    }
84
85    let cli = match Cli::try_parse_from(&raw_args) {
86        Ok(cli) => cli,
87        Err(error) => {
88            if matches!(
89                error.kind(),
90                ErrorKind::DisplayHelp | ErrorKind::DisplayVersion
91            ) {
92                error.exit();
93            }
94            error.exit();
95        }
96    };
97
98    match cli.command.expect("clap requires a command") {
99        Command::Version => print_version(),
100        Command::Skill(args) => match args.command {
101            SkillCommand::Key(key_args) => match key_args.command {
102                SkillKeyCommand::Generate(generate) => commands::skill::run_key_generate(&generate),
103            },
104            SkillCommand::Sign(sign) => commands::skill::run_sign(&sign),
105            SkillCommand::Endorse(endorse) => commands::skill::run_endorse(&endorse),
106            SkillCommand::Verify(verify) => commands::skill::run_verify(&verify),
107            SkillCommand::WhoSigned(who_signed) => {
108                commands::skill::run_who_signed(&who_signed).await
109            }
110            SkillCommand::Trust(trust_args) => match trust_args.command {
111                SkillTrustCommand::Add(add) => commands::skill::run_trust_add(&add),
112                SkillTrustCommand::List(list) => commands::skill::run_trust_list(&list),
113            },
114            SkillCommand::New(new_args) => commands::skills::run_new(&new_args),
115        },
116        Command::Run(args) => {
117            if !args.explain_cost {
118                match (args.eval.as_deref(), args.file.as_deref()) {
119                    (Some(code), None) => {
120                        provider_bootstrap::maybe_seed_ollama_for_inline(
121                            code,
122                            args.yes,
123                            args.llm_mock.is_some(),
124                        )
125                        .await;
126                    }
127                    (None, Some(file)) => {
128                        provider_bootstrap::maybe_seed_ollama_for_run_file(
129                            Path::new(file),
130                            args.yes,
131                            args.llm_mock.is_some(),
132                        )
133                        .await;
134                    }
135                    _ => {}
136                }
137            }
138            let denied =
139                commands::run::build_denied_builtins(args.deny.as_deref(), args.allow.as_deref());
140            let llm_mock_mode = if let Some(path) = args.llm_mock.as_ref() {
141                commands::run::CliLlmMockMode::Replay {
142                    fixture_path: PathBuf::from(path),
143                }
144            } else if let Some(path) = args.llm_mock_record.as_ref() {
145                commands::run::CliLlmMockMode::Record {
146                    fixture_path: PathBuf::from(path),
147                }
148            } else {
149                commands::run::CliLlmMockMode::Off
150            };
151            let attestation = args.attest.then(|| commands::run::RunAttestationOptions {
152                receipt_out: args.receipt_out.as_ref().map(PathBuf::from),
153                agent_id: args.attest_agent.clone(),
154            });
155            let profile_options = run_profile_options(&args.profile);
156
157            match (args.eval.as_deref(), args.file.as_deref()) {
158                (Some(code), None) => {
159                    let (wrapped, tmp) = commands::run::prepare_eval_temp_file(code)
160                        .unwrap_or_else(|e| command_error(&e));
161                    let tmp_path: PathBuf = tmp.path().to_path_buf();
162                    fs::write(&tmp_path, &wrapped).unwrap_or_else(|e| {
163                        command_error(&format!("failed to write temp file for -e: {e}"))
164                    });
165                    let tmp_str = tmp_path.to_string_lossy().into_owned();
166                    if args.explain_cost {
167                        commands::run::run_explain_cost_file_with_skill_dirs(&tmp_str);
168                    } else {
169                        commands::run::run_file_with_skill_dirs(
170                            &tmp_str,
171                            args.trace,
172                            denied,
173                            args.argv.clone(),
174                            args.skill_dir.clone(),
175                            llm_mock_mode.clone(),
176                            attestation.clone(),
177                            profile_options.clone(),
178                        )
179                        .await;
180                    }
181                    drop(tmp);
182                }
183                (None, Some(file)) => {
184                    if args.explain_cost {
185                        commands::run::run_explain_cost_file_with_skill_dirs(file);
186                    } else {
187                        commands::run::run_file_with_skill_dirs(
188                            file,
189                            args.trace,
190                            denied,
191                            args.argv.clone(),
192                            args.skill_dir.clone(),
193                            llm_mock_mode,
194                            attestation,
195                            profile_options,
196                        )
197                        .await
198                    }
199                }
200                (Some(_), Some(_)) => command_error(
201                    "`harn run` accepts either `-e <code>` or `<file.harn>`, not both",
202                ),
203                (None, None) => {
204                    command_error("`harn run` requires either `-e <code>` or `<file.harn>`")
205                }
206            }
207        }
208        Command::Check(args) => {
209            if args.provider_matrix {
210                let cwd = std::env::current_dir().unwrap_or_else(|_| PathBuf::from("."));
211                let extensions = package::load_runtime_extensions(&cwd);
212                package::install_runtime_extensions(&extensions);
213                commands::check::provider_matrix::run(args.format, args.filter.as_deref());
214                return;
215            }
216            if args.connector_matrix {
217                commands::check::connector_matrix::run(
218                    args.format,
219                    args.filter.as_deref(),
220                    &args.targets,
221                );
222                return;
223            }
224            let mut target_strings: Vec<String> = args.targets.clone();
225            if args.workspace {
226                let anchor = target_strings.first().map(Path::new);
227                match package::load_workspace_config(anchor) {
228                    Some((workspace, manifest_dir)) if !workspace.pipelines.is_empty() => {
229                        for pipeline in &workspace.pipelines {
230                            let candidate = Path::new(pipeline);
231                            let resolved = if candidate.is_absolute() {
232                                candidate.to_path_buf()
233                            } else {
234                                manifest_dir.join(candidate)
235                            };
236                            target_strings.push(resolved.to_string_lossy().into_owned());
237                        }
238                    }
239                    Some(_) => command_error(
240                        "--workspace requires `[workspace].pipelines` in the nearest harn.toml",
241                    ),
242                    None => command_error(
243                        "--workspace could not find a harn.toml walking up from the target(s)",
244                    ),
245                }
246            }
247            if target_strings.is_empty() {
248                command_error(
249                    "`harn check` requires at least one target path, or `--workspace` with `[workspace].pipelines`",
250                );
251            }
252            for target in &target_strings {
253                if let Err(error) = package::validate_runtime_manifest_extensions(Path::new(target))
254                {
255                    command_error(&format!("manifest extension validation failed: {error}"));
256                }
257            }
258            let targets: Vec<&str> = target_strings.iter().map(String::as_str).collect();
259            let files = commands::check::collect_harn_targets(&targets);
260            if files.is_empty() {
261                command_error("no .harn files found under the given target(s)");
262            }
263            let module_graph = commands::check::build_module_graph(&files);
264            let cross_file_imports = commands::check::collect_cross_file_imports(&module_graph);
265            let mut should_fail = false;
266            for file in &files {
267                let mut config = package::load_check_config(Some(file));
268                if let Some(path) = args.host_capabilities.as_ref() {
269                    config.host_capabilities_path = Some(path.clone());
270                }
271                if let Some(path) = args.bundle_root.as_ref() {
272                    config.bundle_root = Some(path.clone());
273                }
274                if args.strict_types {
275                    config.strict_types = true;
276                }
277                if let Some(sev) = args.preflight.as_deref() {
278                    config.preflight_severity = Some(sev.to_string());
279                }
280                let outcome = commands::check::check_file_inner(
281                    file,
282                    &config,
283                    &cross_file_imports,
284                    &module_graph,
285                    args.invariants,
286                );
287                should_fail |= outcome.should_fail(config.strict);
288            }
289            if should_fail {
290                process::exit(1);
291            }
292        }
293        Command::Config(args) => {
294            if let Err(error) = commands::config_cmd::run(args).await {
295                command_error(&error);
296            }
297        }
298        Command::Explain(args) => {
299            let code = commands::explain::run_explain(&args);
300            if code != 0 {
301                process::exit(code);
302            }
303        }
304        Command::Contracts(args) => {
305            commands::contracts::handle_contracts_command(args).await;
306        }
307        Command::Connect(args) => {
308            commands::connect::run_connect(*args).await;
309        }
310        Command::Lint(args) => {
311            let targets: Vec<&str> = args.targets.iter().map(String::as_str).collect();
312            let files = commands::check::collect_harn_targets(&targets);
313            if files.is_empty() {
314                command_error("no .harn files found under the given target(s)");
315            }
316            let module_graph = commands::check::build_module_graph(&files);
317            let cross_file_imports = commands::check::collect_cross_file_imports(&module_graph);
318            if args.fix {
319                for file in &files {
320                    let mut config = package::load_check_config(Some(file));
321                    commands::check::apply_harn_lint_config(file, &mut config);
322                    let require_header = args.require_file_header
323                        || commands::check::harn_lint_require_file_header(file);
324                    let complexity_threshold =
325                        commands::check::harn_lint_complexity_threshold(file);
326                    let persona_step_allowlist =
327                        commands::check::harn_lint_persona_step_allowlist(file);
328                    commands::check::lint_fix_file(
329                        file,
330                        &config,
331                        &cross_file_imports,
332                        &module_graph,
333                        require_header,
334                        complexity_threshold,
335                        &persona_step_allowlist,
336                    );
337                }
338            } else {
339                let mut should_fail = false;
340                for file in &files {
341                    let mut config = package::load_check_config(Some(file));
342                    commands::check::apply_harn_lint_config(file, &mut config);
343                    let require_header = args.require_file_header
344                        || commands::check::harn_lint_require_file_header(file);
345                    let complexity_threshold =
346                        commands::check::harn_lint_complexity_threshold(file);
347                    let persona_step_allowlist =
348                        commands::check::harn_lint_persona_step_allowlist(file);
349                    let outcome = commands::check::lint_file_inner(
350                        file,
351                        &config,
352                        &cross_file_imports,
353                        &module_graph,
354                        require_header,
355                        complexity_threshold,
356                        &persona_step_allowlist,
357                    );
358                    should_fail |= outcome.should_fail(config.strict);
359                }
360                if should_fail {
361                    process::exit(1);
362                }
363            }
364        }
365        Command::Fmt(args) => {
366            let targets: Vec<&str> = args.targets.iter().map(String::as_str).collect();
367            // Anchor config resolution on the first target; CLI flags
368            // always win over harn.toml values.
369            let anchor = targets.first().map(Path::new).unwrap_or(Path::new("."));
370            let loaded = match config::load_for_path(anchor) {
371                Ok(c) => c,
372                Err(e) => {
373                    eprintln!("warning: {e}");
374                    config::HarnConfig::default()
375                }
376            };
377            let mut opts = harn_fmt::FmtOptions::default();
378            if let Some(w) = loaded.fmt.line_width {
379                opts.line_width = w;
380            }
381            if let Some(w) = loaded.fmt.separator_width {
382                opts.separator_width = w;
383            }
384            if let Some(w) = args.line_width {
385                opts.line_width = w;
386            }
387            if let Some(w) = args.separator_width {
388                opts.separator_width = w;
389            }
390            commands::check::fmt_targets(
391                &targets,
392                commands::check::FmtMode::from_check_flag(args.check),
393                &opts,
394            );
395        }
396        Command::Test(args) => {
397            if args.target.as_deref() == Some("agents-conformance") {
398                if args.selection.is_some() {
399                    command_error(
400                        "`harn test agents-conformance` does not accept a second positional target; use --category instead",
401                    );
402                }
403                if args.evals || args.determinism || args.record || args.replay || args.watch {
404                    command_error(
405                        "`harn test agents-conformance` cannot be combined with --evals, --determinism, --record, --replay, or --watch",
406                    );
407                }
408                let Some(target_url) = args.agents_target.clone() else {
409                    command_error("`harn test agents-conformance` requires --target <url>");
410                };
411                commands::agents_conformance::run_agents_conformance(
412                    commands::agents_conformance::AgentsConformanceConfig {
413                        target_url,
414                        api_key: args.agents_api_key.clone(),
415                        categories: args.agents_category.clone(),
416                        timeout_ms: args.timeout,
417                        verbose: args.verbose,
418                        json: args.json,
419                        json_out: args.json_out.clone(),
420                        workspace_id: args.agents_workspace_id.clone(),
421                        session_id: args.agents_session_id.clone(),
422                    },
423                )
424                .await;
425                return;
426            }
427            if args.target.as_deref() == Some("protocols") {
428                if args.evals || args.determinism || args.record || args.replay || args.watch {
429                    command_error(
430                        "`harn test protocols` cannot be combined with --evals, --determinism, --record, --replay, or --watch",
431                    );
432                }
433                if args.junit.is_some()
434                    || args.agents_target.is_some()
435                    || args.agents_api_key.is_some()
436                    || !args.agents_category.is_empty()
437                    || args.json
438                    || args.json_out.is_some()
439                    || args.agents_workspace_id.is_some()
440                    || args.agents_session_id.is_some()
441                    || args.parallel
442                    || !args.skill_dir.is_empty()
443                {
444                    command_error(
445                        "`harn test protocols` accepts only --filter, --verbose, --timing, and an optional fixture selection",
446                    );
447                }
448                commands::protocol_conformance::run_protocol_conformance(
449                    args.selection.as_deref(),
450                    args.filter.as_deref(),
451                    args.verbose || args.timing,
452                );
453                return;
454            }
455            if args.evals {
456                if args.determinism || args.record || args.replay || args.watch {
457                    command_error("--evals cannot be combined with --determinism, --record, --replay, or --watch");
458                }
459                if args.target.as_deref() != Some("package") || args.selection.is_some() {
460                    command_error("package evals are run with `harn test package --evals`");
461                }
462                run_package_evals();
463            } else if args.determinism {
464                if args.watch {
465                    command_error("--determinism cannot be combined with --watch");
466                }
467                if args.record || args.replay {
468                    command_error("--determinism manages its own record/replay cycle");
469                }
470                if let Some(t) = args.target.as_deref() {
471                    if t == "conformance" {
472                        commands::test::run_conformance_determinism_tests(
473                            t,
474                            args.selection.as_deref(),
475                            args.filter.as_deref(),
476                            args.timeout,
477                        )
478                        .await;
479                    } else if args.selection.is_some() {
480                        command_error(
481                            "only `harn test conformance` accepts a second positional target",
482                        );
483                    } else {
484                        commands::test::run_determinism_tests(
485                            t,
486                            args.filter.as_deref(),
487                            args.timeout,
488                        )
489                        .await;
490                    }
491                } else {
492                    let test_dir = if PathBuf::from("tests").is_dir() {
493                        "tests".to_string()
494                    } else {
495                        command_error("no path specified and no tests/ directory found");
496                    };
497                    if args.selection.is_some() {
498                        command_error(
499                            "only `harn test conformance` accepts a second positional target",
500                        );
501                    }
502                    commands::test::run_determinism_tests(
503                        &test_dir,
504                        args.filter.as_deref(),
505                        args.timeout,
506                    )
507                    .await;
508                }
509            } else {
510                if args.record {
511                    harn_vm::llm::set_replay_mode(
512                        harn_vm::llm::LlmReplayMode::Record,
513                        ".harn-fixtures",
514                    );
515                } else if args.replay {
516                    harn_vm::llm::set_replay_mode(
517                        harn_vm::llm::LlmReplayMode::Replay,
518                        ".harn-fixtures",
519                    );
520                }
521
522                if let Some(t) = args.target.as_deref() {
523                    if t == "conformance" {
524                        commands::test::run_conformance_tests(
525                            t,
526                            args.selection.as_deref(),
527                            args.filter.as_deref(),
528                            args.junit.as_deref(),
529                            args.timeout,
530                            args.verbose,
531                            args.timing,
532                            args.differential_optimizations,
533                        )
534                        .await;
535                    } else if args.selection.is_some() {
536                        command_error(
537                            "only `harn test conformance` accepts a second positional target",
538                        );
539                    } else if args.watch {
540                        commands::test::run_watch_tests(
541                            t,
542                            args.filter.as_deref(),
543                            args.timeout,
544                            args.parallel,
545                        )
546                        .await;
547                    } else {
548                        commands::test::run_user_tests(
549                            t,
550                            args.filter.as_deref(),
551                            args.timeout,
552                            args.parallel,
553                        )
554                        .await;
555                    }
556                } else {
557                    let test_dir = if PathBuf::from("tests").is_dir() {
558                        "tests".to_string()
559                    } else {
560                        command_error("no path specified and no tests/ directory found");
561                    };
562                    if args.selection.is_some() {
563                        command_error(
564                            "only `harn test conformance` accepts a second positional target",
565                        );
566                    }
567                    if args.watch {
568                        commands::test::run_watch_tests(
569                            &test_dir,
570                            args.filter.as_deref(),
571                            args.timeout,
572                            args.parallel,
573                        )
574                        .await;
575                    } else {
576                        commands::test::run_user_tests(
577                            &test_dir,
578                            args.filter.as_deref(),
579                            args.timeout,
580                            args.parallel,
581                        )
582                        .await;
583                    }
584                }
585            }
586        }
587        Command::Init(args) => commands::init::init_project(args.name.as_deref(), args.template),
588        Command::New(args) => match commands::init::resolve_new_args(&args) {
589            Ok((name, template)) => commands::init::init_project(name.as_deref(), template),
590            Err(error) => {
591                eprintln!("error: {error}");
592                process::exit(1);
593            }
594        },
595        Command::Doctor(args) => {
596            commands::doctor::run_doctor_with_options(commands::doctor::DoctorOptions {
597                network: !args.no_network,
598                json: args.json,
599            })
600            .await
601        }
602        Command::Models(args) => commands::models::run(args).await,
603        Command::Providers(args) => match args.command {
604            ProvidersCommand::Refresh(refresh) => {
605                if let Err(error) = commands::providers::run_refresh(&refresh).await {
606                    command_error(&error);
607                }
608            }
609            ProvidersCommand::Validate(validate) => {
610                if let Err(error) = commands::providers::run_validate(&validate) {
611                    command_error(&error);
612                }
613            }
614            ProvidersCommand::Export(export) => {
615                if let Err(error) = commands::providers::run_export(&export) {
616                    command_error(&error);
617                }
618            }
619        },
620        Command::Try(args) => commands::try_cmd::run(args).await,
621        Command::Quickstart(args) => {
622            if let Err(error) = commands::quickstart::run_quickstart(&args).await {
623                command_error(&error);
624            }
625        }
626        Command::Serve(args) => match args.command {
627            ServeCommand::Acp(args) => {
628                if let Err(error) = commands::serve::run_acp_server(&args).await {
629                    command_error(&error);
630                }
631            }
632            ServeCommand::A2a(args) => {
633                if let Err(error) = commands::serve::run_a2a_server(&args).await {
634                    command_error(&error);
635                }
636            }
637            ServeCommand::Api(args) => {
638                if let Err(error) = commands::serve::run_api_server(&args).await {
639                    command_error(&error);
640                }
641            }
642            ServeCommand::Mcp(args) => {
643                if let Err(error) = commands::serve::run_mcp_server(&args).await {
644                    command_error(&error);
645                }
646            }
647        },
648        Command::Connector(args) => {
649            if let Err(error) = commands::connector::handle_connector_command(args).await {
650                eprintln!("error: {error}");
651                process::exit(1);
652            }
653        }
654        Command::Mcp(args) => commands::mcp::handle_mcp_command(&args.command).await,
655        Command::Watch(args) => {
656            let denied =
657                commands::run::build_denied_builtins(args.deny.as_deref(), args.allow.as_deref());
658            commands::run::run_watch(&args.file, denied).await;
659        }
660        Command::Portal(args) => {
661            commands::portal::run_portal(
662                &args.dir,
663                args.manifest,
664                args.persona_state_dir,
665                &args.host,
666                args.port,
667                args.open,
668                args.allow_remote_launch,
669            )
670            .await
671        }
672        Command::Trigger(args) => {
673            if let Err(error) = commands::trigger::handle(args).await {
674                eprintln!("error: {error}");
675                process::exit(1);
676            }
677        }
678        Command::Flow(args) => match commands::flow::run_flow(&args) {
679            Ok(code) => {
680                if code != 0 {
681                    process::exit(code);
682                }
683            }
684            Err(error) => command_error(&error),
685        },
686        Command::Workflow(args) => match commands::workflow::handle(args) {
687            Ok(code) => {
688                if code != 0 {
689                    process::exit(code);
690                }
691            }
692            Err(error) => command_error(&error),
693        },
694        Command::Supervisor(args) => {
695            if let Err(error) = commands::supervisor::handle(args).await {
696                eprintln!("error: {error}");
697                process::exit(1);
698            }
699        }
700        Command::Trace(args) => {
701            if let Err(error) = commands::trace::handle(args).await {
702                eprintln!("error: {error}");
703                process::exit(1);
704            }
705        }
706        Command::Crystallize(args) => {
707            if let Err(error) = commands::crystallize::run(args) {
708                eprintln!("error: {error}");
709                process::exit(1);
710            }
711        }
712        Command::Trust(args) | Command::TrustGraph(args) => {
713            if let Err(error) = commands::trust::handle(args).await {
714                eprintln!("error: {error}");
715                process::exit(1);
716            }
717        }
718        Command::Verify(args) => {
719            if let Err(error) = verify_provenance_receipt(&args.receipt, args.json) {
720                eprintln!("error: {error}");
721                process::exit(1);
722            }
723        }
724        Command::Completions(args) => print_completions(args.shell),
725        Command::Orchestrator(args) => {
726            if let Err(error) = commands::orchestrator::handle(args).await {
727                eprintln!("error: {error}");
728                process::exit(1);
729            }
730        }
731        Command::Playground(args) => {
732            provider_bootstrap::maybe_seed_ollama_for_playground(
733                Path::new(&args.host),
734                Path::new(&args.script),
735                args.yes,
736                args.llm.is_some(),
737                args.llm_mock.is_some(),
738            )
739            .await;
740            let llm_mock_mode = if let Some(path) = args.llm_mock.as_ref() {
741                commands::run::CliLlmMockMode::Replay {
742                    fixture_path: PathBuf::from(path),
743                }
744            } else if let Some(path) = args.llm_mock_record.as_ref() {
745                commands::run::CliLlmMockMode::Record {
746                    fixture_path: PathBuf::from(path),
747                }
748            } else {
749                commands::run::CliLlmMockMode::Off
750            };
751            if let Err(error) = commands::playground::run_command(args, llm_mock_mode).await {
752                eprint!("{error}");
753                process::exit(1);
754            }
755        }
756        Command::Runs(args) => match args.command {
757            RunsCommand::Inspect(inspect) => {
758                inspect_run_record(&inspect.path, inspect.compare.as_deref())
759            }
760        },
761        Command::Session(args) => commands::session::run(args),
762        Command::Replay(args) => replay_run_record(&args.path),
763        Command::Eval(args) => {
764            let llm_mock_mode = if let Some(path) = args.llm_mock.as_ref() {
765                commands::run::CliLlmMockMode::Replay {
766                    fixture_path: PathBuf::from(path),
767                }
768            } else if let Some(path) = args.llm_mock_record.as_ref() {
769                commands::run::CliLlmMockMode::Record {
770                    fixture_path: PathBuf::from(path),
771                }
772            } else {
773                commands::run::CliLlmMockMode::Off
774            };
775            eval_run_record(
776                &args.path,
777                args.compare.as_deref(),
778                args.structural_experiment.as_deref(),
779                &args.argv,
780                &llm_mock_mode,
781            )
782        }
783        Command::Repl => commands::repl::run_repl().await,
784        Command::Bench(args) => commands::bench::run(args).await,
785        Command::TestBench(args) => commands::test_bench::run(args.command).await,
786        Command::Viz(args) => commands::viz::run_viz(&args.file, args.output.as_deref()),
787        Command::Install(args) => package::install_packages(
788            args.frozen || args.locked || args.offline,
789            args.refetch.as_deref(),
790            args.offline,
791            args.json,
792        ),
793        Command::Add(args) => package::add_package_with_registry(
794            &args.name_or_spec,
795            args.alias.as_deref(),
796            args.git.as_deref(),
797            args.tag.as_deref(),
798            args.rev.as_deref(),
799            args.branch.as_deref(),
800            args.path.as_deref(),
801            args.registry.as_deref(),
802        ),
803        Command::Update(args) => {
804            package::update_packages(args.alias.as_deref(), args.all, args.json)
805        }
806        Command::Remove(args) => package::remove_package(&args.alias),
807        Command::Lock => package::lock_packages(),
808        Command::Package(args) => match args.command {
809            PackageCommand::List(list) => package::list_packages(list.json),
810            PackageCommand::Doctor(doctor) => package::doctor_packages(doctor.json),
811            PackageCommand::Search(search) => package::search_package_registry(
812                search.query.as_deref(),
813                search.registry.as_deref(),
814                search.json,
815            ),
816            PackageCommand::Info(info) => {
817                package::show_package_registry_info(&info.name, info.registry.as_deref(), info.json)
818            }
819            PackageCommand::Check(check) => {
820                package::check_package(check.package.as_deref(), check.json)
821            }
822            PackageCommand::Pack(pack) => package::pack_package(
823                pack.package.as_deref(),
824                pack.output.as_deref(),
825                pack.dry_run,
826                pack.json,
827            ),
828            PackageCommand::Docs(docs) => package::generate_package_docs(
829                docs.package.as_deref(),
830                docs.output.as_deref(),
831                docs.check,
832            ),
833            PackageCommand::Cache(cache) => match cache.command {
834                PackageCacheCommand::List => package::list_package_cache(),
835                PackageCacheCommand::Clean(clean) => package::clean_package_cache(clean.all),
836                PackageCacheCommand::Verify(verify) => {
837                    package::verify_package_cache(verify.materialized)
838                }
839            },
840            PackageCommand::Outdated(args) => package::outdated_packages(
841                args.refresh,
842                args.remote,
843                args.registry.as_deref(),
844                args.json,
845            ),
846            PackageCommand::Audit(args) => {
847                package::audit_packages(args.registry.as_deref(), args.skip_materialized, args.json)
848            }
849            PackageCommand::Artifacts(args) => match args.command {
850                PackageArtifactsCommand::Manifest(manifest) => {
851                    package::artifacts_manifest(manifest.output.as_deref())
852                }
853                PackageArtifactsCommand::Check(check) => {
854                    package::artifacts_check(&check.manifest, check.json)
855                }
856            },
857        },
858        Command::Publish(args) => package::publish_package(
859            args.package.as_deref(),
860            args.dry_run,
861            args.registry.as_deref(),
862            args.json,
863        ),
864        Command::MergeCaptain(args) => match args.command {
865            MergeCaptainCommand::Run(run) => {
866                let code = commands::merge_captain::run_driver(&run);
867                if code != 0 {
868                    process::exit(code);
869                }
870            }
871            MergeCaptainCommand::Ladder(ladder) => {
872                let code = commands::merge_captain::run_ladder(&ladder);
873                if code != 0 {
874                    process::exit(code);
875                }
876            }
877            MergeCaptainCommand::Iterate(iterate) => {
878                let code = commands::merge_captain::run_iterate(&iterate);
879                if code != 0 {
880                    process::exit(code);
881                }
882            }
883            MergeCaptainCommand::Audit(audit) => {
884                let code = commands::merge_captain::run_audit(&audit);
885                if code != 0 {
886                    process::exit(code);
887                }
888            }
889            MergeCaptainCommand::Mock(mock) => {
890                let code = match mock {
891                    MergeCaptainMockCommand::Init(args) => {
892                        commands::merge_captain_mock::run_init(&args)
893                    }
894                    MergeCaptainMockCommand::Step(args) => {
895                        commands::merge_captain_mock::run_step(&args)
896                    }
897                    MergeCaptainMockCommand::Status(args) => {
898                        commands::merge_captain_mock::run_status(&args)
899                    }
900                    MergeCaptainMockCommand::Serve(args) => {
901                        commands::merge_captain_mock::run_serve(&args).await
902                    }
903                    MergeCaptainMockCommand::Cleanup(args) => {
904                        commands::merge_captain_mock::run_cleanup(&args)
905                    }
906                    MergeCaptainMockCommand::Scenarios => {
907                        commands::merge_captain_mock::run_scenarios()
908                    }
909                };
910                if code != 0 {
911                    process::exit(code);
912                }
913            }
914        },
915        Command::Persona(args) => match args.command {
916            PersonaCommand::New(new) => {
917                if let Err(error) = commands::persona_scaffold::run_new(&new) {
918                    eprintln!("error: {error}");
919                    process::exit(1);
920                }
921            }
922            PersonaCommand::Doctor(doctor) => {
923                if let Err(error) =
924                    commands::persona_doctor::run_doctor(args.manifest.as_deref(), &doctor).await
925                {
926                    eprintln!("error: {error}");
927                    process::exit(1);
928                }
929            }
930            PersonaCommand::Check(check) => {
931                commands::persona::run_check(args.manifest.as_deref(), &check)
932            }
933            PersonaCommand::List(list) => {
934                commands::persona::run_list(args.manifest.as_deref(), &list)
935            }
936            PersonaCommand::Inspect(inspect) => {
937                commands::persona::run_inspect(args.manifest.as_deref(), &inspect)
938            }
939            PersonaCommand::Status(status) => {
940                if let Err(error) = commands::persona::run_status(
941                    args.manifest.as_deref(),
942                    &args.state_dir,
943                    &status,
944                )
945                .await
946                {
947                    eprintln!("error: {error}");
948                    process::exit(1);
949                }
950            }
951            PersonaCommand::Pause(control) => {
952                if let Err(error) = commands::persona::run_pause(
953                    args.manifest.as_deref(),
954                    &args.state_dir,
955                    &control,
956                )
957                .await
958                {
959                    eprintln!("error: {error}");
960                    process::exit(1);
961                }
962            }
963            PersonaCommand::Resume(control) => {
964                if let Err(error) = commands::persona::run_resume(
965                    args.manifest.as_deref(),
966                    &args.state_dir,
967                    &control,
968                )
969                .await
970                {
971                    eprintln!("error: {error}");
972                    process::exit(1);
973                }
974            }
975            PersonaCommand::Disable(control) => {
976                if let Err(error) = commands::persona::run_disable(
977                    args.manifest.as_deref(),
978                    &args.state_dir,
979                    &control,
980                )
981                .await
982                {
983                    eprintln!("error: {error}");
984                    process::exit(1);
985                }
986            }
987            PersonaCommand::Tick(tick) => {
988                if let Err(error) =
989                    commands::persona::run_tick(args.manifest.as_deref(), &args.state_dir, &tick)
990                        .await
991                {
992                    eprintln!("error: {error}");
993                    process::exit(1);
994                }
995            }
996            PersonaCommand::Trigger(trigger) => {
997                if let Err(error) = commands::persona::run_trigger(
998                    args.manifest.as_deref(),
999                    &args.state_dir,
1000                    &trigger,
1001                )
1002                .await
1003                {
1004                    eprintln!("error: {error}");
1005                    process::exit(1);
1006                }
1007            }
1008            PersonaCommand::Spend(spend) => {
1009                if let Err(error) =
1010                    commands::persona::run_spend(args.manifest.as_deref(), &args.state_dir, &spend)
1011                        .await
1012                {
1013                    eprintln!("error: {error}");
1014                    process::exit(1);
1015                }
1016            }
1017            PersonaCommand::Supervision(supervision) => match supervision.command {
1018                PersonaSupervisionCommand::Tail(tail) => {
1019                    if let Err(error) = commands::persona_supervision::run_tail(
1020                        args.manifest.as_deref(),
1021                        &args.state_dir,
1022                        &tail,
1023                    )
1024                    .await
1025                    {
1026                        eprintln!("error: {error}");
1027                        process::exit(1);
1028                    }
1029                }
1030            },
1031        },
1032        Command::ModelInfo(args) => {
1033            if !print_model_info(&args).await {
1034                process::exit(1);
1035            }
1036        }
1037        Command::ProviderCatalog(args) => print_provider_catalog(args.available_only),
1038        Command::ProviderReady(args) => {
1039            run_provider_ready(
1040                &args.provider,
1041                args.model.as_deref(),
1042                args.base_url.as_deref(),
1043                args.json,
1044            )
1045            .await
1046        }
1047        Command::Skills(args) => match args.command {
1048            SkillsCommand::List(list) => commands::skills::run_list(&list),
1049            SkillsCommand::Inspect(inspect) => commands::skills::run_inspect(&inspect),
1050            SkillsCommand::Match(matcher) => commands::skills::run_match(&matcher),
1051            SkillsCommand::Install(install) => commands::skills::run_install(&install),
1052            SkillsCommand::New(new_args) => commands::skills::run_new(&new_args),
1053        },
1054        Command::Tool(args) => match args.command {
1055            ToolCommand::New(new_args) => {
1056                if let Err(error) = commands::tool::run_new(&new_args) {
1057                    eprintln!("error: {error}");
1058                    process::exit(1);
1059                }
1060            }
1061        },
1062        Command::DumpHighlightKeywords(args) => {
1063            commands::dump_highlight_keywords::run(&args.output, args.check);
1064        }
1065        Command::DumpTriggerQuickref(args) => {
1066            commands::dump_trigger_quickref::run(&args.output, args.check);
1067        }
1068        Command::DumpConnectorMatrix(args) => {
1069            commands::check::connector_matrix::run_docs(&args.output, &args.sources, args.check);
1070        }
1071        Command::DumpProtocolArtifacts(args) => {
1072            commands::dump_protocol_artifacts::run(&args.output_dir, args.check);
1073        }
1074    }
1075}
1076
1077fn run_profile_options(args: &cli::ProfileArgs) -> commands::run::RunProfileOptions {
1078    commands::run::RunProfileOptions {
1079        text: args.text,
1080        json_path: args.json_path.clone(),
1081    }
1082}
1083
1084fn print_completions(shell: CompletionShell) {
1085    let mut command = Cli::command();
1086    let shell = clap_complete::Shell::from(shell);
1087    clap_complete::generate(shell, &mut command, "harn", &mut std::io::stdout());
1088}
1089
1090fn normalize_serve_args(mut raw_args: Vec<String>) -> Vec<String> {
1091    if raw_args.len() > 2
1092        && raw_args.get(1).is_some_and(|arg| arg == "serve")
1093        && !matches!(
1094            raw_args.get(2).map(String::as_str),
1095            Some("acp" | "a2a" | "api" | "mcp" | "-h" | "--help")
1096        )
1097    {
1098        raw_args.insert(2, "a2a".to_string());
1099    }
1100    raw_args
1101}
1102
1103fn print_version() {
1104    println!(
1105        r#"
1106 ╱▔▔╲
1107 ╱    ╲    harn v{}
1108 │ ◆  │    the agent harness language
1109 │    │
1110 ╰──╯╱
1111   ╱╱
1112"#,
1113        env!("CARGO_PKG_VERSION")
1114    );
1115}
1116
1117async fn print_model_info(args: &ModelInfoArgs) -> bool {
1118    let resolved = harn_vm::llm_config::resolve_model_info(&args.model);
1119    let api_key_result = harn_vm::llm::resolve_api_key(&resolved.provider);
1120    let api_key_set = api_key_result.is_ok();
1121    let api_key = api_key_result.unwrap_or_default();
1122    let context_window =
1123        harn_vm::llm::fetch_provider_max_context(&resolved.provider, &resolved.id, &api_key).await;
1124    let readiness = local_openai_readiness(&resolved.provider, &resolved.id, &api_key).await;
1125    let catalog = harn_vm::llm_config::model_catalog_entry(&resolved.id);
1126    let runtime_context_window = catalog
1127        .as_ref()
1128        .and_then(|entry| entry.runtime_context_window);
1129    let capabilities = harn_vm::llm::capabilities::lookup(&resolved.provider, &resolved.id);
1130    let mut payload = serde_json::json!({
1131        "alias": args.model,
1132        "id": resolved.id,
1133        "provider": resolved.provider,
1134        "resolved_alias": resolved.alias,
1135        "tool_format": resolved.tool_format,
1136        "tier": resolved.tier,
1137        "api_key_set": api_key_set,
1138        "context_window": context_window,
1139        "runtime_context_window": runtime_context_window,
1140        "readiness": readiness,
1141        "catalog": catalog,
1142        "capabilities": {
1143            "native_tools": capabilities.native_tools,
1144            "defer_loading": capabilities.defer_loading,
1145            "tool_search": capabilities.tool_search,
1146            "max_tools": capabilities.max_tools,
1147            "prompt_caching": capabilities.prompt_caching,
1148            "vision": capabilities.vision,
1149            "vision_supported": capabilities.vision_supported,
1150            "audio": capabilities.audio,
1151            "pdf": capabilities.pdf,
1152            "files_api_supported": capabilities.files_api_supported,
1153            "json_schema": capabilities.json_schema,
1154            "thinking": !capabilities.thinking_modes.is_empty(),
1155            "thinking_modes": capabilities.thinking_modes,
1156            "interleaved_thinking_supported": capabilities.interleaved_thinking_supported,
1157            "anthropic_beta_features": capabilities.anthropic_beta_features,
1158            "preserve_thinking": capabilities.preserve_thinking,
1159            "server_parser": capabilities.server_parser,
1160            "honors_chat_template_kwargs": capabilities.honors_chat_template_kwargs,
1161            "recommended_endpoint": capabilities.recommended_endpoint,
1162            "text_tool_wire_format_supported": capabilities.text_tool_wire_format_supported,
1163        },
1164        "qc_default_model": harn_vm::llm_config::qc_default_model(&resolved.provider),
1165    });
1166
1167    let should_verify = args.verify || args.warm;
1168    let mut ok = true;
1169    if should_verify {
1170        if resolved.provider == "ollama" {
1171            let mut readiness = harn_vm::llm::OllamaReadinessOptions::new(resolved.id.clone());
1172            readiness.warm = args.warm;
1173            readiness.observe_loaded = true;
1174            readiness.keep_alive = args
1175                .keep_alive
1176                .as_deref()
1177                .and_then(harn_vm::llm::normalize_ollama_keep_alive);
1178            let result = harn_vm::llm::ollama_readiness(readiness).await;
1179            ok = result.valid;
1180            payload["readiness"] = serde_json::to_value(&result).unwrap_or_else(|error| {
1181                serde_json::json!({
1182                    "valid": false,
1183                    "status": "serialization_error",
1184                    "message": format!("failed to serialize readiness result: {error}"),
1185                })
1186            });
1187        } else {
1188            ok = false;
1189            payload["readiness"] = serde_json::json!({
1190                "valid": false,
1191                "status": "unsupported_provider",
1192                "message": format!(
1193                    "model-info --verify is only supported for Ollama models; resolved provider is '{}'",
1194                    resolved.provider
1195                ),
1196                "provider": resolved.provider,
1197            });
1198        }
1199    }
1200
1201    println!(
1202        "{}",
1203        serde_json::to_string(&payload).unwrap_or_else(|error| {
1204            command_error(&format!("failed to serialize model info: {error}"))
1205        })
1206    );
1207    ok
1208}
1209
1210async fn local_openai_readiness(
1211    provider: &str,
1212    model: &str,
1213    api_key: &str,
1214) -> Option<serde_json::Value> {
1215    let def = harn_vm::llm_config::provider_config(provider)?;
1216    if def.auth_style != "none" || !harn_vm::llm::supports_model_readiness_probe(&def) {
1217        return None;
1218    }
1219    let readiness = harn_vm::llm::probe_openai_compatible_model(provider, model, api_key).await;
1220    Some(serde_json::json!({
1221        "valid": readiness.valid,
1222        "category": readiness.category,
1223        "message": readiness.message,
1224        "provider": readiness.provider,
1225        "model": readiness.model,
1226        "url": readiness.url,
1227        "status": readiness.status,
1228        "available_models": readiness.available_models,
1229    }))
1230}
1231
1232fn print_provider_catalog(available_only: bool) {
1233    let provider_names = if available_only {
1234        harn_vm::llm_config::available_provider_names()
1235    } else {
1236        harn_vm::llm_config::provider_names()
1237    };
1238    let providers: Vec<_> = provider_names
1239        .into_iter()
1240        .filter_map(|name| {
1241            harn_vm::llm_config::provider_config(&name).map(|def| {
1242                serde_json::json!({
1243                    "name": name,
1244                    "display_name": def.display_name,
1245                    "icon": def.icon,
1246                    "base_url": harn_vm::llm_config::resolve_base_url(&def),
1247                    "base_url_env": def.base_url_env,
1248                    "auth_style": def.auth_style,
1249                    "auth_envs": harn_vm::llm_config::auth_env_names(&def.auth_env),
1250                    "auth_available": harn_vm::llm_config::provider_key_available(&name),
1251                    "features": def.features,
1252                    "cost_per_1k_in": def.cost_per_1k_in,
1253                    "cost_per_1k_out": def.cost_per_1k_out,
1254                    "latency_p50_ms": def.latency_p50_ms,
1255                })
1256            })
1257        })
1258        .collect();
1259    let models: Vec<_> = harn_vm::llm_config::model_catalog_entries()
1260        .into_iter()
1261        .map(|(id, model)| {
1262            serde_json::json!({
1263                "id": id,
1264                "name": model.name,
1265                "provider": model.provider,
1266                "context_window": model.context_window,
1267                "runtime_context_window": model.runtime_context_window,
1268                "stream_timeout": model.stream_timeout,
1269                "capabilities": model.capabilities,
1270                "pricing": model.pricing,
1271            })
1272        })
1273        .collect();
1274    let aliases: Vec<_> = harn_vm::llm_config::alias_entries()
1275        .into_iter()
1276        .map(|(name, alias)| {
1277            serde_json::json!({
1278                "name": name,
1279                "id": alias.id,
1280                "provider": alias.provider,
1281                "tool_format": alias.tool_format,
1282            })
1283        })
1284        .collect();
1285    let payload = serde_json::json!({
1286        "providers": providers,
1287        "known_model_names": harn_vm::llm_config::known_model_names(),
1288        "available_providers": harn_vm::llm_config::available_provider_names(),
1289        "aliases": aliases,
1290        "models": models,
1291        "qc_defaults": harn_vm::llm_config::qc_defaults(),
1292    });
1293    println!(
1294        "{}",
1295        serde_json::to_string(&payload).unwrap_or_else(|error| {
1296            command_error(&format!("failed to serialize provider catalog: {error}"))
1297        })
1298    );
1299}
1300
1301async fn run_provider_ready(
1302    provider: &str,
1303    model: Option<&str>,
1304    base_url: Option<&str>,
1305    json: bool,
1306) {
1307    let readiness =
1308        harn_vm::llm::readiness::probe_provider_readiness(provider, model, base_url).await;
1309    if json {
1310        match serde_json::to_string_pretty(&readiness) {
1311            Ok(payload) => println!("{payload}"),
1312            Err(error) => command_error(&format!("failed to serialize readiness result: {error}")),
1313        }
1314    } else if readiness.ok {
1315        println!("{}", readiness.message);
1316    } else {
1317        eprintln!("{}", readiness.message);
1318    }
1319    if !readiness.ok {
1320        process::exit(1);
1321    }
1322}
1323
1324fn command_error(message: &str) -> ! {
1325    Cli::command()
1326        .error(ErrorKind::ValueValidation, message)
1327        .exit()
1328}
1329
1330fn verify_provenance_receipt(path: &str, json: bool) -> Result<(), String> {
1331    let raw =
1332        fs::read_to_string(path).map_err(|error| format!("failed to read {path}: {error}"))?;
1333    let receipt: harn_vm::ProvenanceReceipt = serde_json::from_str(&raw)
1334        .map_err(|error| format!("failed to parse provenance receipt {path}: {error}"))?;
1335    let report = harn_vm::verify_receipt(&receipt);
1336    if json {
1337        println!(
1338            "{}",
1339            serde_json::to_string_pretty(&report).map_err(|error| error.to_string())?
1340        );
1341    } else if report.verified {
1342        println!(
1343            "verified receipt={} events={} receipt_hash={} event_root_hash={}",
1344            report.receipt_id.unwrap_or_else(|| "-".to_string()),
1345            report.event_count,
1346            report.receipt_hash.unwrap_or_else(|| "-".to_string()),
1347            report.event_root_hash.unwrap_or_else(|| "-".to_string())
1348        );
1349    } else {
1350        println!(
1351            "failed receipt={} events={}",
1352            report.receipt_id.unwrap_or_else(|| "-".to_string()),
1353            report.event_count
1354        );
1355        for error in &report.errors {
1356            println!("  {error}");
1357        }
1358        return Err("provenance receipt verification failed".to_string());
1359    }
1360    Ok(())
1361}
1362
1363fn load_run_record_or_exit(path: &Path) -> harn_vm::orchestration::RunRecord {
1364    match harn_vm::orchestration::load_run_record(path) {
1365        Ok(run) => run,
1366        Err(error) => {
1367            eprintln!("Failed to load run record: {error}");
1368            process::exit(1);
1369        }
1370    }
1371}
1372
1373fn load_eval_suite_manifest_or_exit(path: &Path) -> harn_vm::orchestration::EvalSuiteManifest {
1374    harn_vm::orchestration::load_eval_suite_manifest(path).unwrap_or_else(|error| {
1375        eprintln!("Failed to load eval manifest {}: {error}", path.display());
1376        process::exit(1);
1377    })
1378}
1379
1380fn load_eval_pack_manifest_or_exit(path: &Path) -> harn_vm::orchestration::EvalPackManifest {
1381    harn_vm::orchestration::load_eval_pack_manifest(path).unwrap_or_else(|error| {
1382        eprintln!("Failed to load eval pack {}: {error}", path.display());
1383        process::exit(1);
1384    })
1385}
1386
1387fn load_persona_eval_ladder_manifest_or_exit(
1388    path: &Path,
1389) -> harn_vm::orchestration::PersonaEvalLadderManifest {
1390    harn_vm::orchestration::load_persona_eval_ladder_manifest(path).unwrap_or_else(|error| {
1391        eprintln!(
1392            "Failed to load persona eval ladder {}: {error}",
1393            path.display()
1394        );
1395        process::exit(1);
1396    })
1397}
1398
1399fn file_looks_like_eval_manifest(path: &Path) -> bool {
1400    if path.file_name().and_then(|name| name.to_str()) == Some("harn.eval.toml") {
1401        return true;
1402    }
1403    if path.extension().and_then(|ext| ext.to_str()) == Some("toml") {
1404        let Ok(content) = fs::read_to_string(path) else {
1405            return false;
1406        };
1407        return toml::from_str::<harn_vm::orchestration::EvalPackManifest>(&content)
1408            .is_ok_and(|manifest| !manifest.cases.is_empty() || !manifest.ladders.is_empty());
1409    }
1410    let Ok(content) = fs::read_to_string(path) else {
1411        return false;
1412    };
1413    let Ok(json) = serde_json::from_str::<serde_json::Value>(&content) else {
1414        return false;
1415    };
1416    json.get("_type").and_then(|value| value.as_str()) == Some("eval_suite_manifest")
1417        || json.get("cases").is_some()
1418}
1419
1420fn file_looks_like_eval_pack_manifest(path: &Path) -> bool {
1421    if path.file_name().and_then(|name| name.to_str()) == Some("harn.eval.toml") {
1422        return true;
1423    }
1424    if path.extension().and_then(|ext| ext.to_str()) == Some("toml") {
1425        return file_looks_like_eval_manifest(path);
1426    }
1427    let Ok(content) = fs::read_to_string(path) else {
1428        return false;
1429    };
1430    let Ok(json) = serde_json::from_str::<serde_json::Value>(&content) else {
1431        return false;
1432    };
1433    json.get("version").is_some()
1434        && (json.get("cases").is_some() || json.get("ladders").is_some())
1435        && json.get("_type").and_then(|value| value.as_str()) != Some("eval_suite_manifest")
1436}
1437
1438fn file_looks_like_persona_eval_ladder_manifest(path: &Path) -> bool {
1439    let Ok(content) = fs::read_to_string(path) else {
1440        return false;
1441    };
1442    if path.extension().and_then(|ext| ext.to_str()) == Some("json") {
1443        let Ok(json) = serde_json::from_str::<serde_json::Value>(&content) else {
1444            return false;
1445        };
1446        return json.get("_type").and_then(|value| value.as_str())
1447            == Some("persona_eval_ladder_manifest")
1448            || json.get("timeout_tiers").is_some()
1449            || json.get("timeout-tiers").is_some();
1450    }
1451    toml::from_str::<harn_vm::orchestration::PersonaEvalLadderManifest>(&content).is_ok_and(
1452        |manifest| {
1453            manifest
1454                .type_name
1455                .eq_ignore_ascii_case("persona_eval_ladder_manifest")
1456                || (!manifest.timeout_tiers.is_empty() && manifest.backend.path.is_some())
1457        },
1458    )
1459}
1460
1461fn collect_run_record_paths(path: &str) -> Vec<PathBuf> {
1462    let path = Path::new(path);
1463    if path.is_file() {
1464        return vec![path.to_path_buf()];
1465    }
1466    if path.is_dir() {
1467        let mut entries: Vec<PathBuf> = fs::read_dir(path)
1468            .unwrap_or_else(|error| {
1469                eprintln!("Failed to read run directory {}: {error}", path.display());
1470                process::exit(1);
1471            })
1472            .filter_map(|entry| entry.ok().map(|entry| entry.path()))
1473            .filter(|entry| entry.extension().and_then(|ext| ext.to_str()) == Some("json"))
1474            .collect();
1475        entries.sort();
1476        return entries;
1477    }
1478    eprintln!("Run path does not exist: {}", path.display());
1479    process::exit(1);
1480}
1481
1482fn print_run_diff(diff: &harn_vm::orchestration::RunDiffReport) {
1483    println!(
1484        "Diff: {} -> {} [{} -> {}]",
1485        diff.left_run_id, diff.right_run_id, diff.left_status, diff.right_status
1486    );
1487    println!("Identical: {}", diff.identical);
1488    println!("Stage diffs: {}", diff.stage_diffs.len());
1489    println!("Tool diffs: {}", diff.tool_diffs.len());
1490    println!("Observability diffs: {}", diff.observability_diffs.len());
1491    println!("Transition delta: {}", diff.transition_count_delta);
1492    println!("Artifact delta: {}", diff.artifact_count_delta);
1493    println!("Checkpoint delta: {}", diff.checkpoint_count_delta);
1494    for stage in &diff.stage_diffs {
1495        println!("- {} [{}]", stage.node_id, stage.change);
1496        for detail in &stage.details {
1497            println!("  {}", detail);
1498        }
1499    }
1500    for tool in &diff.tool_diffs {
1501        println!("- tool {} [{}]", tool.tool_name, tool.args_hash);
1502        println!("  left: {:?}", tool.left_result);
1503        println!("  right: {:?}", tool.right_result);
1504    }
1505    for item in &diff.observability_diffs {
1506        println!("- {} [{}]", item.label, item.section);
1507        for detail in &item.details {
1508            println!("  {}", detail);
1509        }
1510    }
1511}
1512
1513fn inspect_run_record(path: &str, compare: Option<&str>) {
1514    let run = load_run_record_or_exit(Path::new(path));
1515    println!("Run: {}", run.id);
1516    println!(
1517        "Workflow: {}",
1518        run.workflow_name
1519            .clone()
1520            .unwrap_or_else(|| run.workflow_id.clone())
1521    );
1522    println!("Status: {}", run.status);
1523    println!("Task: {}", run.task);
1524    println!("Stages: {}", run.stages.len());
1525    println!("Artifacts: {}", run.artifacts.len());
1526    println!("Transitions: {}", run.transitions.len());
1527    println!("Checkpoints: {}", run.checkpoints.len());
1528    println!("HITL questions: {}", run.hitl_questions.len());
1529    if let Some(observability) = &run.observability {
1530        println!("Planner rounds: {}", observability.planner_rounds.len());
1531        println!("Research facts: {}", observability.research_fact_count);
1532        println!("Workers: {}", observability.worker_lineage.len());
1533        println!(
1534            "Action graph: {} nodes / {} edges",
1535            observability.action_graph_nodes.len(),
1536            observability.action_graph_edges.len()
1537        );
1538        println!(
1539            "Transcript pointers: {}",
1540            observability.transcript_pointers.len()
1541        );
1542        println!("Daemon events: {}", observability.daemon_events.len());
1543    }
1544    if let Some(parent_worker_id) = run
1545        .metadata
1546        .get("parent_worker_id")
1547        .and_then(|value| value.as_str())
1548    {
1549        println!("Parent worker: {}", parent_worker_id);
1550    }
1551    if let Some(parent_stage_id) = run
1552        .metadata
1553        .get("parent_stage_id")
1554        .and_then(|value| value.as_str())
1555    {
1556        println!("Parent stage: {}", parent_stage_id);
1557    }
1558    if run
1559        .metadata
1560        .get("delegated")
1561        .and_then(|value| value.as_bool())
1562        .unwrap_or(false)
1563    {
1564        println!("Delegated: true");
1565    }
1566    println!(
1567        "Pending nodes: {}",
1568        if run.pending_nodes.is_empty() {
1569            "-".to_string()
1570        } else {
1571            run.pending_nodes.join(", ")
1572        }
1573    );
1574    println!(
1575        "Replay fixture: {}",
1576        if run.replay_fixture.is_some() {
1577            "embedded"
1578        } else {
1579            "derived"
1580        }
1581    );
1582    for stage in &run.stages {
1583        let worker = stage.metadata.get("worker");
1584        let worker_suffix = worker
1585            .and_then(|value| value.get("name"))
1586            .and_then(|value| value.as_str())
1587            .map(|name| format!(" worker={name}"))
1588            .unwrap_or_default();
1589        println!(
1590            "- {} [{}] status={} outcome={} branch={}{}",
1591            stage.node_id,
1592            stage.kind,
1593            stage.status,
1594            stage.outcome,
1595            stage.branch.clone().unwrap_or_else(|| "-".to_string()),
1596            worker_suffix,
1597        );
1598        if let Some(worker) = worker {
1599            if let Some(worker_id) = worker.get("id").and_then(|value| value.as_str()) {
1600                println!("  worker_id: {}", worker_id);
1601            }
1602            if let Some(child_run_id) = worker.get("child_run_id").and_then(|value| value.as_str())
1603            {
1604                println!("  child_run_id: {}", child_run_id);
1605            }
1606            if let Some(child_run_path) = worker
1607                .get("child_run_path")
1608                .and_then(|value| value.as_str())
1609            {
1610                println!("  child_run_path: {}", child_run_path);
1611            }
1612        }
1613    }
1614    if let Some(observability) = &run.observability {
1615        for round in &observability.planner_rounds {
1616            println!(
1617                "- planner {} iterations={} llm_calls={} tools={} research_facts={}",
1618                round.node_id,
1619                round.iteration_count,
1620                round.llm_call_count,
1621                round.tool_execution_count,
1622                round.research_facts.len()
1623            );
1624        }
1625        for pointer in &observability.transcript_pointers {
1626            println!(
1627                "- transcript {} [{}] available={} {}",
1628                pointer.label,
1629                pointer.kind,
1630                pointer.available,
1631                pointer
1632                    .path
1633                    .clone()
1634                    .unwrap_or_else(|| pointer.location.clone())
1635            );
1636        }
1637        for event in &observability.daemon_events {
1638            println!(
1639                "- daemon {} [{:?}] at {}",
1640                event.name, event.kind, event.timestamp
1641            );
1642            println!("  id: {}", event.daemon_id);
1643            println!("  persist_path: {}", event.persist_path);
1644            if let Some(summary) = &event.payload_summary {
1645                println!("  payload: {}", summary);
1646            }
1647        }
1648    }
1649    if let Some(compare_path) = compare {
1650        let baseline = load_run_record_or_exit(Path::new(compare_path));
1651        print_run_diff(&harn_vm::orchestration::diff_run_records(&baseline, &run));
1652    }
1653}
1654
1655fn replay_run_record(path: &str) {
1656    let run = load_run_record_or_exit(Path::new(path));
1657    println!("Replay: {}", run.id);
1658    for stage in &run.stages {
1659        println!(
1660            "[{}] status={} outcome={} branch={}",
1661            stage.node_id,
1662            stage.status,
1663            stage.outcome,
1664            stage.branch.clone().unwrap_or_else(|| "-".to_string())
1665        );
1666        if let Some(text) = &stage.visible_text {
1667            println!("  visible: {}", text);
1668        }
1669        if let Some(verification) = &stage.verification {
1670            println!("  verification: {}", verification);
1671        }
1672    }
1673    if let Some(transcript) = &run.transcript {
1674        println!(
1675            "Transcript events persisted: {}",
1676            transcript["events"]
1677                .as_array()
1678                .map(|v| v.len())
1679                .unwrap_or(0)
1680        );
1681    }
1682    let fixture = run
1683        .replay_fixture
1684        .clone()
1685        .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(&run));
1686    let report = harn_vm::orchestration::evaluate_run_against_fixture(&run, &fixture);
1687    println!(
1688        "Embedded replay fixture: {}",
1689        if report.pass { "PASS" } else { "FAIL" }
1690    );
1691    for transition in &run.transitions {
1692        println!(
1693            "transition {} -> {} ({})",
1694            transition
1695                .from_node_id
1696                .clone()
1697                .unwrap_or_else(|| "start".to_string()),
1698            transition.to_node_id,
1699            transition
1700                .branch
1701                .clone()
1702                .unwrap_or_else(|| "default".to_string())
1703        );
1704    }
1705}
1706
1707fn eval_run_record(
1708    path: &str,
1709    compare: Option<&str>,
1710    structural_experiment: Option<&str>,
1711    argv: &[String],
1712    llm_mock_mode: &commands::run::CliLlmMockMode,
1713) {
1714    if let Some(experiment) = structural_experiment {
1715        let path_buf = PathBuf::from(path);
1716        if !path_buf.is_file() || path_buf.extension().and_then(|ext| ext.to_str()) != Some("harn")
1717        {
1718            eprintln!(
1719                "--structural-experiment currently requires a .harn pipeline path, got {}",
1720                path
1721            );
1722            process::exit(1);
1723        }
1724        if compare.is_some() {
1725            eprintln!("--compare cannot be combined with --structural-experiment");
1726            process::exit(1);
1727        }
1728        if matches!(llm_mock_mode, commands::run::CliLlmMockMode::Record { .. }) {
1729            eprintln!("--llm-mock-record cannot be combined with --structural-experiment");
1730            process::exit(1);
1731        }
1732        let path_buf = fs::canonicalize(&path_buf).unwrap_or_else(|error| {
1733            command_error(&format!(
1734                "failed to canonicalize structural eval pipeline {}: {error}",
1735                path_buf.display()
1736            ))
1737        });
1738        run_structural_experiment_eval(&path_buf, experiment, argv, llm_mock_mode);
1739        return;
1740    }
1741
1742    let path_buf = PathBuf::from(path);
1743    if path_buf.is_file() && file_looks_like_persona_eval_ladder_manifest(&path_buf) {
1744        if compare.is_some() {
1745            eprintln!("--compare is not supported with persona eval ladder manifests");
1746            process::exit(1);
1747        }
1748        let manifest = load_persona_eval_ladder_manifest_or_exit(&path_buf);
1749        let report =
1750            harn_vm::orchestration::run_persona_eval_ladder(&manifest).unwrap_or_else(|error| {
1751                eprintln!(
1752                    "Failed to evaluate persona eval ladder {}: {error}",
1753                    path_buf.display()
1754                );
1755                process::exit(1);
1756            });
1757        print_persona_ladder_report(&report);
1758        if !report.pass {
1759            process::exit(1);
1760        }
1761        return;
1762    }
1763
1764    if path_buf.is_file() && file_looks_like_eval_pack_manifest(&path_buf) {
1765        if compare.is_some() {
1766            eprintln!("--compare is not supported with eval pack manifests");
1767            process::exit(1);
1768        }
1769        let manifest = load_eval_pack_manifest_or_exit(&path_buf);
1770        let report = harn_vm::orchestration::evaluate_eval_pack_manifest(&manifest).unwrap_or_else(
1771            |error| {
1772                eprintln!(
1773                    "Failed to evaluate eval pack {}: {error}",
1774                    path_buf.display()
1775                );
1776                process::exit(1);
1777            },
1778        );
1779        print_eval_pack_report(&report);
1780        if !report.pass {
1781            process::exit(1);
1782        }
1783        return;
1784    }
1785
1786    if path_buf.is_file() && file_looks_like_eval_manifest(&path_buf) {
1787        if compare.is_some() {
1788            eprintln!("--compare is not supported with eval suite manifests");
1789            process::exit(1);
1790        }
1791        let manifest = load_eval_suite_manifest_or_exit(&path_buf);
1792        let suite = harn_vm::orchestration::evaluate_run_suite_manifest(&manifest).unwrap_or_else(
1793            |error| {
1794                eprintln!(
1795                    "Failed to evaluate manifest {}: {error}",
1796                    path_buf.display()
1797                );
1798                process::exit(1);
1799            },
1800        );
1801        println!(
1802            "{} {} passed, {} failed, {} total",
1803            if suite.pass { "PASS" } else { "FAIL" },
1804            suite.passed,
1805            suite.failed,
1806            suite.total
1807        );
1808        for case in &suite.cases {
1809            println!(
1810                "- {} [{}] {}",
1811                case.label.clone().unwrap_or_else(|| case.run_id.clone()),
1812                case.workflow_id,
1813                if case.pass { "PASS" } else { "FAIL" }
1814            );
1815            if let Some(path) = &case.source_path {
1816                println!("  path: {}", path);
1817            }
1818            if let Some(comparison) = &case.comparison {
1819                println!("  baseline identical: {}", comparison.identical);
1820                if !comparison.identical {
1821                    println!(
1822                        "  baseline status: {} -> {}",
1823                        comparison.left_status, comparison.right_status
1824                    );
1825                }
1826            }
1827            for failure in &case.failures {
1828                println!("  {}", failure);
1829            }
1830        }
1831        if !suite.pass {
1832            process::exit(1);
1833        }
1834        return;
1835    }
1836
1837    let paths = collect_run_record_paths(path);
1838    if paths.len() > 1 {
1839        let mut cases = Vec::new();
1840        for path in &paths {
1841            let run = load_run_record_or_exit(path);
1842            let fixture = run
1843                .replay_fixture
1844                .clone()
1845                .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(&run));
1846            cases.push((run, fixture, Some(path.display().to_string())));
1847        }
1848        let suite = harn_vm::orchestration::evaluate_run_suite(cases);
1849        println!(
1850            "{} {} passed, {} failed, {} total",
1851            if suite.pass { "PASS" } else { "FAIL" },
1852            suite.passed,
1853            suite.failed,
1854            suite.total
1855        );
1856        for case in &suite.cases {
1857            println!(
1858                "- {} [{}] {}",
1859                case.run_id,
1860                case.workflow_id,
1861                if case.pass { "PASS" } else { "FAIL" }
1862            );
1863            if let Some(path) = &case.source_path {
1864                println!("  path: {}", path);
1865            }
1866            if let Some(comparison) = &case.comparison {
1867                println!("  baseline identical: {}", comparison.identical);
1868            }
1869            for failure in &case.failures {
1870                println!("  {}", failure);
1871            }
1872        }
1873        if !suite.pass {
1874            process::exit(1);
1875        }
1876        return;
1877    }
1878
1879    let run = load_run_record_or_exit(&paths[0]);
1880    let fixture = run
1881        .replay_fixture
1882        .clone()
1883        .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(&run));
1884    let report = harn_vm::orchestration::evaluate_run_against_fixture(&run, &fixture);
1885    println!("{}", if report.pass { "PASS" } else { "FAIL" });
1886    println!("Stages: {}", report.stage_count);
1887    if let Some(compare_path) = compare {
1888        let baseline = load_run_record_or_exit(Path::new(compare_path));
1889        print_run_diff(&harn_vm::orchestration::diff_run_records(&baseline, &run));
1890    }
1891    if !report.failures.is_empty() {
1892        for failure in &report.failures {
1893            println!("- {}", failure);
1894        }
1895    }
1896    if !report.pass {
1897        process::exit(1);
1898    }
1899}
1900
1901fn print_eval_pack_report(report: &harn_vm::orchestration::EvalPackReport) {
1902    println!(
1903        "{} {} passed, {} blocking failed, {} warning, {} informational, {} total",
1904        if report.pass { "PASS" } else { "FAIL" },
1905        report.passed,
1906        report.blocking_failed,
1907        report.warning_failed,
1908        report.informational_failed,
1909        report.total
1910    );
1911    for case in &report.cases {
1912        println!(
1913            "- {} [{}] {} ({})",
1914            case.label,
1915            case.workflow_id,
1916            if case.pass { "PASS" } else { "FAIL" },
1917            case.severity
1918        );
1919        if let Some(path) = &case.source_path {
1920            println!("  path: {}", path);
1921        }
1922        if let Some(comparison) = &case.comparison {
1923            println!("  baseline identical: {}", comparison.identical);
1924            if !comparison.identical {
1925                println!(
1926                    "  baseline status: {} -> {}",
1927                    comparison.left_status, comparison.right_status
1928                );
1929            }
1930        }
1931        for failure in &case.failures {
1932            println!("  {}", failure);
1933        }
1934        for warning in &case.warnings {
1935            println!("  warning: {}", warning);
1936        }
1937        for item in &case.informational {
1938            println!("  info: {}", item);
1939        }
1940    }
1941    for ladder in &report.ladders {
1942        println!(
1943            "- ladder {} [{}] {} ({}) first_correct={}/{}",
1944            ladder.id,
1945            ladder.persona,
1946            if ladder.pass { "PASS" } else { "FAIL" },
1947            ladder.severity,
1948            ladder.first_correct_route.as_deref().unwrap_or("<none>"),
1949            ladder.first_correct_tier.as_deref().unwrap_or("<none>")
1950        );
1951        println!("  artifacts: {}", ladder.artifact_root);
1952        for tier in &ladder.tiers {
1953            println!(
1954                "  - {} [{}] {} tools={} models={} latency={}ms cost=${:.6}",
1955                tier.timeout_tier,
1956                tier.route_id,
1957                tier.outcome,
1958                tier.tool_calls,
1959                tier.model_calls,
1960                tier.latency_ms,
1961                tier.cost_usd
1962            );
1963            for reason in &tier.degradation_reasons {
1964                println!("    {}", reason);
1965            }
1966        }
1967    }
1968}
1969
1970fn print_persona_ladder_report(report: &harn_vm::orchestration::PersonaEvalLadderReport) {
1971    println!(
1972        "{} ladder {} passed, {} degraded/looped, {} total",
1973        if report.pass { "PASS" } else { "FAIL" },
1974        report.passed,
1975        report.failed,
1976        report.total
1977    );
1978    println!(
1979        "first_correct: {}/{}",
1980        report.first_correct_route.as_deref().unwrap_or("<none>"),
1981        report.first_correct_tier.as_deref().unwrap_or("<none>")
1982    );
1983    println!("artifacts: {}", report.artifact_root);
1984    for tier in &report.tiers {
1985        println!(
1986            "- {} [{}] {} tools={} models={} latency={}ms cost=${:.6}",
1987            tier.timeout_tier,
1988            tier.route_id,
1989            tier.outcome,
1990            tier.tool_calls,
1991            tier.model_calls,
1992            tier.latency_ms,
1993            tier.cost_usd
1994        );
1995        for reason in &tier.degradation_reasons {
1996            println!("  {}", reason);
1997        }
1998    }
1999}
2000
2001fn run_package_evals() {
2002    let paths = package::load_package_eval_pack_paths(None).unwrap_or_else(|error| {
2003        eprintln!("{error}");
2004        process::exit(1);
2005    });
2006    let mut all_pass = true;
2007    for path in &paths {
2008        println!("Eval pack: {}", path.display());
2009        let manifest = load_eval_pack_manifest_or_exit(path);
2010        let report = harn_vm::orchestration::evaluate_eval_pack_manifest(&manifest).unwrap_or_else(
2011            |error| {
2012                eprintln!("Failed to evaluate eval pack {}: {error}", path.display());
2013                process::exit(1);
2014            },
2015        );
2016        print_eval_pack_report(&report);
2017        all_pass &= report.pass;
2018    }
2019    if !all_pass {
2020        process::exit(1);
2021    }
2022}
2023
2024fn run_structural_experiment_eval(
2025    path: &Path,
2026    experiment: &str,
2027    argv: &[String],
2028    llm_mock_mode: &commands::run::CliLlmMockMode,
2029) {
2030    let baseline_dir = tempfile::Builder::new()
2031        .prefix("harn-eval-baseline-")
2032        .tempdir()
2033        .unwrap_or_else(|error| {
2034            command_error(&format!("failed to create baseline tempdir: {error}"))
2035        });
2036    let variant_dir = tempfile::Builder::new()
2037        .prefix("harn-eval-variant-")
2038        .tempdir()
2039        .unwrap_or_else(|error| {
2040            command_error(&format!("failed to create variant tempdir: {error}"))
2041        });
2042
2043    let baseline = spawn_eval_pipeline_run(path, baseline_dir.path(), None, argv, llm_mock_mode);
2044    if !baseline.status.success() {
2045        relay_subprocess_failure("baseline", &baseline);
2046    }
2047
2048    let variant = spawn_eval_pipeline_run(
2049        path,
2050        variant_dir.path(),
2051        Some(experiment),
2052        argv,
2053        llm_mock_mode,
2054    );
2055    if !variant.status.success() {
2056        relay_subprocess_failure("variant", &variant);
2057    }
2058
2059    let baseline_runs = collect_structural_eval_runs(baseline_dir.path());
2060    let variant_runs = collect_structural_eval_runs(variant_dir.path());
2061    if baseline_runs.is_empty() || variant_runs.is_empty() {
2062        eprintln!(
2063            "structural eval expected workflow run records under {} and {}, but one side was empty",
2064            baseline_dir.path().display(),
2065            variant_dir.path().display()
2066        );
2067        process::exit(1);
2068    }
2069    if baseline_runs.len() != variant_runs.len() {
2070        eprintln!(
2071            "structural eval produced different run counts: baseline={} variant={}",
2072            baseline_runs.len(),
2073            variant_runs.len()
2074        );
2075        process::exit(1);
2076    }
2077
2078    let mut baseline_ok = 0usize;
2079    let mut variant_ok = 0usize;
2080    let mut any_failures = false;
2081
2082    println!("Structural experiment: {}", experiment);
2083    println!("Cases: {}", baseline_runs.len());
2084    for (baseline_run, variant_run) in baseline_runs.iter().zip(variant_runs.iter()) {
2085        let baseline_fixture = baseline_run
2086            .replay_fixture
2087            .clone()
2088            .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(baseline_run));
2089        let variant_fixture = variant_run
2090            .replay_fixture
2091            .clone()
2092            .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(variant_run));
2093        let baseline_report =
2094            harn_vm::orchestration::evaluate_run_against_fixture(baseline_run, &baseline_fixture);
2095        let variant_report =
2096            harn_vm::orchestration::evaluate_run_against_fixture(variant_run, &variant_fixture);
2097        let diff = harn_vm::orchestration::diff_run_records(baseline_run, variant_run);
2098        if baseline_report.pass {
2099            baseline_ok += 1;
2100        }
2101        if variant_report.pass {
2102            variant_ok += 1;
2103        }
2104        any_failures |= !baseline_report.pass || !variant_report.pass;
2105        println!(
2106            "- {} [{}]",
2107            variant_run
2108                .workflow_name
2109                .clone()
2110                .unwrap_or_else(|| variant_run.workflow_id.clone()),
2111            variant_run.task
2112        );
2113        println!(
2114            "  baseline: {}",
2115            if baseline_report.pass { "PASS" } else { "FAIL" }
2116        );
2117        for failure in &baseline_report.failures {
2118            println!("    {}", failure);
2119        }
2120        println!(
2121            "  variant: {}",
2122            if variant_report.pass { "PASS" } else { "FAIL" }
2123        );
2124        for failure in &variant_report.failures {
2125            println!("    {}", failure);
2126        }
2127        println!("  diff identical: {}", diff.identical);
2128        println!("  stage diffs: {}", diff.stage_diffs.len());
2129        println!("  tool diffs: {}", diff.tool_diffs.len());
2130        println!("  observability diffs: {}", diff.observability_diffs.len());
2131    }
2132
2133    println!("Baseline {} / {} passed", baseline_ok, baseline_runs.len());
2134    println!("Variant {} / {} passed", variant_ok, variant_runs.len());
2135
2136    if any_failures {
2137        process::exit(1);
2138    }
2139}
2140
2141fn spawn_eval_pipeline_run(
2142    path: &Path,
2143    run_dir: &Path,
2144    structural_experiment: Option<&str>,
2145    argv: &[String],
2146    llm_mock_mode: &commands::run::CliLlmMockMode,
2147) -> std::process::Output {
2148    let exe = env::current_exe().unwrap_or_else(|error| {
2149        command_error(&format!("failed to resolve current executable: {error}"))
2150    });
2151    let mut command = std::process::Command::new(exe);
2152    command.current_dir(path.parent().unwrap_or_else(|| Path::new(".")));
2153    command.arg("run");
2154    match llm_mock_mode {
2155        commands::run::CliLlmMockMode::Off => {}
2156        commands::run::CliLlmMockMode::Replay { fixture_path } => {
2157            command
2158                .arg("--llm-mock")
2159                .arg(absolute_cli_path(fixture_path));
2160        }
2161        commands::run::CliLlmMockMode::Record { fixture_path } => {
2162            command
2163                .arg("--llm-mock-record")
2164                .arg(absolute_cli_path(fixture_path));
2165        }
2166    }
2167    command.arg(path);
2168    if !argv.is_empty() {
2169        command.arg("--");
2170        command.args(argv);
2171    }
2172    command.env(harn_vm::runtime_paths::HARN_RUN_DIR_ENV, run_dir);
2173    if let Some(experiment) = structural_experiment {
2174        command.env("HARN_STRUCTURAL_EXPERIMENT", experiment);
2175    }
2176    command.output().unwrap_or_else(|error| {
2177        command_error(&format!(
2178            "failed to spawn `harn run {}` for structural eval: {error}",
2179            path.display()
2180        ))
2181    })
2182}
2183
2184fn absolute_cli_path(path: &Path) -> PathBuf {
2185    if path.is_absolute() {
2186        return path.to_path_buf();
2187    }
2188    env::current_dir()
2189        .unwrap_or_else(|_| PathBuf::from("."))
2190        .join(path)
2191}
2192
2193fn relay_subprocess_failure(label: &str, output: &std::process::Output) -> ! {
2194    let stdout = String::from_utf8_lossy(&output.stdout);
2195    let stderr = String::from_utf8_lossy(&output.stderr);
2196    if !stdout.trim().is_empty() {
2197        eprintln!("[{label}] stdout:\n{stdout}");
2198    }
2199    if !stderr.trim().is_empty() {
2200        eprintln!("[{label}] stderr:\n{stderr}");
2201    }
2202    process::exit(output.status.code().unwrap_or(1));
2203}
2204
2205fn collect_structural_eval_runs(dir: &Path) -> Vec<harn_vm::orchestration::RunRecord> {
2206    let mut paths: Vec<PathBuf> = fs::read_dir(dir)
2207        .unwrap_or_else(|error| {
2208            command_error(&format!(
2209                "failed to read structural eval run dir {}: {error}",
2210                dir.display()
2211            ))
2212        })
2213        .filter_map(|entry| entry.ok().map(|entry| entry.path()))
2214        .filter(|entry| entry.extension().and_then(|ext| ext.to_str()) == Some("json"))
2215        .collect();
2216    paths.sort();
2217    let mut runs: Vec<_> = paths
2218        .iter()
2219        .map(|path| load_run_record_or_exit(path))
2220        .collect();
2221    runs.sort_by(|left, right| {
2222        (
2223            left.started_at.as_str(),
2224            left.workflow_id.as_str(),
2225            left.task.as_str(),
2226        )
2227            .cmp(&(
2228                right.started_at.as_str(),
2229                right.workflow_id.as_str(),
2230                right.task.as_str(),
2231            ))
2232    });
2233    runs
2234}
2235
2236/// Parse a .harn file, returning (source, AST). Exits on error.
2237pub(crate) fn parse_source_file(path: &str) -> (String, Vec<harn_parser::SNode>) {
2238    let source = match fs::read_to_string(path) {
2239        Ok(s) => s,
2240        Err(e) => {
2241            eprintln!("Error reading {path}: {e}");
2242            process::exit(1);
2243        }
2244    };
2245
2246    let mut lexer = Lexer::new(&source);
2247    let tokens = match lexer.tokenize() {
2248        Ok(t) => t,
2249        Err(e) => {
2250            let diagnostic = harn_parser::diagnostic::render_diagnostic(
2251                &source,
2252                path,
2253                &error_span_from_lex(&e),
2254                "error",
2255                &e.to_string(),
2256                Some("here"),
2257                None,
2258            );
2259            eprint!("{diagnostic}");
2260            process::exit(1);
2261        }
2262    };
2263
2264    let mut parser = Parser::new(tokens);
2265    let program = match parser.parse() {
2266        Ok(p) => p,
2267        Err(err) => {
2268            if parser.all_errors().is_empty() {
2269                let span = error_span_from_parse(&err);
2270                let diagnostic = harn_parser::diagnostic::render_diagnostic(
2271                    &source,
2272                    path,
2273                    &span,
2274                    "error",
2275                    &harn_parser::diagnostic::parser_error_message(&err),
2276                    Some(harn_parser::diagnostic::parser_error_label(&err)),
2277                    harn_parser::diagnostic::parser_error_help(&err),
2278                );
2279                eprint!("{diagnostic}");
2280            } else {
2281                for e in parser.all_errors() {
2282                    let span = error_span_from_parse(e);
2283                    let diagnostic = harn_parser::diagnostic::render_diagnostic(
2284                        &source,
2285                        path,
2286                        &span,
2287                        "error",
2288                        &harn_parser::diagnostic::parser_error_message(e),
2289                        Some(harn_parser::diagnostic::parser_error_label(e)),
2290                        harn_parser::diagnostic::parser_error_help(e),
2291                    );
2292                    eprint!("{diagnostic}");
2293                }
2294            }
2295            process::exit(1);
2296        }
2297    };
2298
2299    (source, program)
2300}
2301
2302fn error_span_from_lex(e: &harn_lexer::LexerError) -> harn_lexer::Span {
2303    match e {
2304        harn_lexer::LexerError::UnexpectedCharacter(_, span)
2305        | harn_lexer::LexerError::UnterminatedString(span)
2306        | harn_lexer::LexerError::UnterminatedBlockComment(span) => *span,
2307    }
2308}
2309
2310fn error_span_from_parse(e: &harn_parser::ParserError) -> harn_lexer::Span {
2311    match e {
2312        harn_parser::ParserError::Unexpected { span, .. } => *span,
2313        harn_parser::ParserError::UnexpectedEof { span, .. } => *span,
2314    }
2315}
2316
2317/// Execute source code and return the output. Used by REPL and conformance tests.
2318pub(crate) async fn execute(source: &str, source_path: Option<&Path>) -> Result<String, String> {
2319    let mut lexer = Lexer::new(source);
2320    let tokens = lexer.tokenize().map_err(|e| e.to_string())?;
2321    let mut parser = Parser::new(tokens);
2322    let program = parser.parse().map_err(|e| e.to_string())?;
2323
2324    // Static cross-module resolution: when executed from a file, derive the
2325    // import graph so `execute` catches undefined calls at typecheck time.
2326    // The REPL / `-e` path invokes this without `source_path`, where there
2327    // is no importing file context; we fall back to no-imports checking.
2328    let mut checker = TypeChecker::new();
2329    if let Some(path) = source_path {
2330        let graph = harn_modules::build(&[path.to_path_buf()]);
2331        if let Some(imported) = graph.imported_names_for_file(path) {
2332            checker = checker.with_imported_names(imported);
2333        }
2334        if let Some(imported) = graph.imported_type_declarations_for_file(path) {
2335            checker = checker.with_imported_type_decls(imported);
2336        }
2337        if let Some(imported) = graph.imported_callable_declarations_for_file(path) {
2338            checker = checker.with_imported_callable_decls(imported);
2339        }
2340    }
2341    let type_diagnostics = checker.check(&program);
2342    let mut warning_lines = Vec::new();
2343    for diag in &type_diagnostics {
2344        match diag.severity {
2345            DiagnosticSeverity::Error => return Err(diag.message.clone()),
2346            DiagnosticSeverity::Warning => {
2347                warning_lines.push(format!("warning: {}", diag.message));
2348            }
2349        }
2350    }
2351
2352    let chunk = harn_vm::Compiler::new()
2353        .compile(&program)
2354        .map_err(|e| e.to_string())?;
2355
2356    let local = tokio::task::LocalSet::new();
2357    local
2358        .run_until(async {
2359            let mut vm = harn_vm::Vm::new();
2360            harn_vm::register_vm_stdlib(&mut vm);
2361            install_default_hostlib(&mut vm);
2362            let source_parent = source_path
2363                .and_then(|p| p.parent())
2364                .unwrap_or(std::path::Path::new("."));
2365            let project_root = harn_vm::stdlib::process::find_project_root(source_parent);
2366            let store_base = project_root.as_deref().unwrap_or(source_parent);
2367            let execution_cwd = std::env::current_dir()
2368                .unwrap_or_else(|_| std::path::PathBuf::from("."))
2369                .to_string_lossy()
2370                .into_owned();
2371            let source_dir = source_parent.to_string_lossy().into_owned();
2372            if source_path.is_some_and(is_conformance_path) {
2373                harn_vm::event_log::install_memory_for_current_thread(64);
2374            }
2375            harn_vm::register_store_builtins(&mut vm, store_base);
2376            harn_vm::register_metadata_builtins(&mut vm, store_base);
2377            let pipeline_name = source_path
2378                .and_then(|p| p.file_stem())
2379                .and_then(|s| s.to_str())
2380                .unwrap_or("default");
2381            harn_vm::register_checkpoint_builtins(&mut vm, store_base, pipeline_name);
2382            harn_vm::stdlib::process::set_thread_execution_context(Some(
2383                harn_vm::orchestration::RunExecutionRecord {
2384                    cwd: Some(execution_cwd),
2385                    source_dir: Some(source_dir),
2386                    env: std::collections::BTreeMap::new(),
2387                    adapter: None,
2388                    repo_path: None,
2389                    worktree_path: None,
2390                    branch: None,
2391                    base_ref: None,
2392                    cleanup: None,
2393                },
2394            ));
2395            if let Some(ref root) = project_root {
2396                vm.set_project_root(root);
2397            }
2398            if let Some(path) = source_path {
2399                if let Some(parent) = path.parent() {
2400                    if !parent.as_os_str().is_empty() {
2401                        vm.set_source_dir(parent);
2402                    }
2403                }
2404            }
2405            // Conformance tests land here via `run_conformance_tests`; for
2406            // `skill_fs_*` fixtures to see the bundled `skills/` folder
2407            // we run the same layered discovery as `harn run`.
2408            let loaded = skill_loader::load_skills(&skill_loader::SkillLoaderInputs {
2409                cli_dirs: Vec::new(),
2410                source_path: source_path.map(Path::to_path_buf),
2411            });
2412            skill_loader::emit_loader_warnings(&loaded.loader_warnings);
2413            skill_loader::install_skills_global(&mut vm, &loaded);
2414            if let Some(path) = source_path {
2415                let extensions = package::load_runtime_extensions(path);
2416                package::install_runtime_extensions(&extensions);
2417                package::install_manifest_triggers(&mut vm, &extensions)
2418                    .await
2419                    .map_err(|error| format!("failed to install manifest triggers: {error}"))?;
2420                package::install_manifest_hooks(&mut vm, &extensions)
2421                    .await
2422                    .map_err(|error| format!("failed to install manifest hooks: {error}"))?;
2423            }
2424            let _event_log = harn_vm::event_log::active_event_log()
2425                .unwrap_or_else(|| harn_vm::event_log::install_memory_for_current_thread(64));
2426            let connector_clients_installed =
2427                should_install_default_connector_clients(source, source_path);
2428            if connector_clients_installed {
2429                install_default_connector_clients(store_base)
2430                    .await
2431                    .map_err(|error| format!("failed to initialize connector clients: {error}"))?;
2432            }
2433            let execution_result = vm.execute(&chunk).await.map_err(|e| e.to_string());
2434            harn_vm::egress::reset_egress_policy_for_host();
2435            if connector_clients_installed {
2436                harn_vm::clear_active_connector_clients();
2437            }
2438            harn_vm::stdlib::process::set_thread_execution_context(None);
2439            execution_result?;
2440            let mut output = String::new();
2441            for wl in &warning_lines {
2442                output.push_str(wl);
2443                output.push('\n');
2444            }
2445            output.push_str(vm.output());
2446            Ok(output)
2447        })
2448        .await
2449}
2450
2451fn should_install_default_connector_clients(source: &str, source_path: Option<&Path>) -> bool {
2452    if !source_path.is_some_and(is_conformance_path) {
2453        return true;
2454    }
2455    source.contains("connector_call")
2456        || source.contains("std/connectors")
2457        || source.contains("connectors/")
2458}
2459
2460fn is_conformance_path(path: &Path) -> bool {
2461    path.components()
2462        .any(|component| component.as_os_str() == "conformance")
2463}
2464
2465async fn install_default_connector_clients(base_dir: &Path) -> Result<(), String> {
2466    let event_log = harn_vm::event_log::active_event_log()
2467        .unwrap_or_else(|| harn_vm::event_log::install_memory_for_current_thread(64));
2468    let secret_namespace = connector_secret_namespace(base_dir);
2469    let secrets: Arc<dyn harn_vm::secrets::SecretProvider> = Arc::new(
2470        harn_vm::secrets::configured_default_chain(secret_namespace)
2471            .map_err(|error| format!("failed to configure secret providers: {error}"))?,
2472    );
2473
2474    let registry = harn_vm::ConnectorRegistry::default();
2475    let metrics = Arc::new(harn_vm::MetricsRegistry::default());
2476    let inbox = Arc::new(
2477        harn_vm::InboxIndex::new(event_log.clone(), metrics.clone())
2478            .await
2479            .map_err(|error| error.to_string())?,
2480    );
2481    registry
2482        .init_all(harn_vm::ConnectorCtx {
2483            event_log,
2484            secrets,
2485            inbox,
2486            metrics,
2487            rate_limiter: Arc::new(harn_vm::RateLimiterFactory::default()),
2488        })
2489        .await
2490        .map_err(|error| error.to_string())?;
2491    let clients = registry.client_map().await;
2492    harn_vm::install_active_connector_clients(clients);
2493    Ok(())
2494}
2495
2496fn connector_secret_namespace(base_dir: &Path) -> String {
2497    match std::env::var("HARN_SECRET_NAMESPACE") {
2498        Ok(namespace) if !namespace.trim().is_empty() => namespace,
2499        _ => {
2500            let leaf = base_dir
2501                .file_name()
2502                .and_then(|name| name.to_str())
2503                .filter(|name| !name.is_empty())
2504                .unwrap_or("workspace");
2505            format!("harn/{leaf}")
2506        }
2507    }
2508}
2509
2510#[cfg(test)]
2511mod main_tests {
2512    use super::{normalize_serve_args, should_install_default_connector_clients};
2513    use std::path::Path;
2514
2515    #[test]
2516    fn normalize_serve_args_inserts_a2a_for_legacy_shape() {
2517        let args = normalize_serve_args(vec![
2518            "harn".to_string(),
2519            "serve".to_string(),
2520            "--port".to_string(),
2521            "3000".to_string(),
2522            "agent.harn".to_string(),
2523        ]);
2524        assert_eq!(
2525            args,
2526            vec![
2527                "harn".to_string(),
2528                "serve".to_string(),
2529                "a2a".to_string(),
2530                "--port".to_string(),
2531                "3000".to_string(),
2532                "agent.harn".to_string(),
2533            ]
2534        );
2535    }
2536
2537    #[test]
2538    fn normalize_serve_args_preserves_explicit_subcommands() {
2539        let args = normalize_serve_args(vec![
2540            "harn".to_string(),
2541            "serve".to_string(),
2542            "acp".to_string(),
2543            "server.harn".to_string(),
2544        ]);
2545        assert_eq!(
2546            args,
2547            vec![
2548                "harn".to_string(),
2549                "serve".to_string(),
2550                "acp".to_string(),
2551                "server.harn".to_string(),
2552            ]
2553        );
2554    }
2555
2556    #[test]
2557    fn conformance_skips_connector_clients_unless_fixture_uses_connectors() {
2558        let path = Path::new("conformance/tests/language/basic.harn");
2559        assert!(!should_install_default_connector_clients(
2560            "println(1)",
2561            Some(path)
2562        ));
2563        assert!(!should_install_default_connector_clients(
2564            "trust_graph_verify_chain()",
2565            Some(path)
2566        ));
2567        assert!(should_install_default_connector_clients(
2568            "import { post_message } from \"std/connectors/slack\"",
2569            Some(path)
2570        ));
2571        assert!(should_install_default_connector_clients(
2572            "println(1)",
2573            Some(Path::new("examples/demo.harn"))
2574        ));
2575    }
2576}