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