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                            args.differential_optimizations,
529                        )
530                        .await;
531                    } else if args.selection.is_some() {
532                        command_error(
533                            "only `harn test conformance` accepts a second positional target",
534                        );
535                    } else if args.watch {
536                        commands::test::run_watch_tests(
537                            t,
538                            args.filter.as_deref(),
539                            args.timeout,
540                            args.parallel,
541                        )
542                        .await;
543                    } else {
544                        commands::test::run_user_tests(
545                            t,
546                            args.filter.as_deref(),
547                            args.timeout,
548                            args.parallel,
549                        )
550                        .await;
551                    }
552                } else {
553                    let test_dir = if PathBuf::from("tests").is_dir() {
554                        "tests".to_string()
555                    } else {
556                        command_error("no path specified and no tests/ directory found");
557                    };
558                    if args.selection.is_some() {
559                        command_error(
560                            "only `harn test conformance` accepts a second positional target",
561                        );
562                    }
563                    if args.watch {
564                        commands::test::run_watch_tests(
565                            &test_dir,
566                            args.filter.as_deref(),
567                            args.timeout,
568                            args.parallel,
569                        )
570                        .await;
571                    } else {
572                        commands::test::run_user_tests(
573                            &test_dir,
574                            args.filter.as_deref(),
575                            args.timeout,
576                            args.parallel,
577                        )
578                        .await;
579                    }
580                }
581            }
582        }
583        Command::Init(args) => commands::init::init_project(args.name.as_deref(), args.template),
584        Command::New(args) => match commands::init::resolve_new_args(&args) {
585            Ok((name, template)) => commands::init::init_project(name.as_deref(), template),
586            Err(error) => {
587                eprintln!("error: {error}");
588                process::exit(1);
589            }
590        },
591        Command::Doctor(args) => {
592            commands::doctor::run_doctor_with_options(commands::doctor::DoctorOptions {
593                network: !args.no_network,
594                json: args.json,
595            })
596            .await
597        }
598        Command::Models(args) => commands::models::run(args).await,
599        Command::Try(args) => commands::try_cmd::run(args).await,
600        Command::Quickstart(args) => {
601            if let Err(error) = commands::quickstart::run_quickstart(&args).await {
602                command_error(&error);
603            }
604        }
605        Command::Serve(args) => match args.command {
606            ServeCommand::Acp(args) => {
607                if let Err(error) = commands::serve::run_acp_server(&args).await {
608                    command_error(&error);
609                }
610            }
611            ServeCommand::A2a(args) => {
612                if let Err(error) = commands::serve::run_a2a_server(&args).await {
613                    command_error(&error);
614                }
615            }
616            ServeCommand::Mcp(args) => {
617                if let Err(error) = commands::serve::run_mcp_server(&args).await {
618                    command_error(&error);
619                }
620            }
621        },
622        Command::Connector(args) => {
623            if let Err(error) = commands::connector::handle_connector_command(args).await {
624                eprintln!("error: {error}");
625                process::exit(1);
626            }
627        }
628        Command::Mcp(args) => commands::mcp::handle_mcp_command(&args.command).await,
629        Command::Watch(args) => {
630            let denied =
631                commands::run::build_denied_builtins(args.deny.as_deref(), args.allow.as_deref());
632            commands::run::run_watch(&args.file, denied).await;
633        }
634        Command::Portal(args) => {
635            commands::portal::run_portal(
636                &args.dir,
637                args.manifest,
638                args.persona_state_dir,
639                &args.host,
640                args.port,
641                args.open,
642            )
643            .await
644        }
645        Command::Trigger(args) => {
646            if let Err(error) = commands::trigger::handle(args).await {
647                eprintln!("error: {error}");
648                process::exit(1);
649            }
650        }
651        Command::Flow(args) => match commands::flow::run_flow(&args) {
652            Ok(code) => {
653                if code != 0 {
654                    process::exit(code);
655                }
656            }
657            Err(error) => command_error(&error),
658        },
659        Command::Workflow(args) => match commands::workflow::handle(args) {
660            Ok(code) => {
661                if code != 0 {
662                    process::exit(code);
663                }
664            }
665            Err(error) => command_error(&error),
666        },
667        Command::Supervisor(args) => {
668            if let Err(error) = commands::supervisor::handle(args).await {
669                eprintln!("error: {error}");
670                process::exit(1);
671            }
672        }
673        Command::Trace(args) => {
674            if let Err(error) = commands::trace::handle(args).await {
675                eprintln!("error: {error}");
676                process::exit(1);
677            }
678        }
679        Command::Crystallize(args) => {
680            if let Err(error) = commands::crystallize::run(args) {
681                eprintln!("error: {error}");
682                process::exit(1);
683            }
684        }
685        Command::Trust(args) | Command::TrustGraph(args) => {
686            if let Err(error) = commands::trust::handle(args).await {
687                eprintln!("error: {error}");
688                process::exit(1);
689            }
690        }
691        Command::Verify(args) => {
692            if let Err(error) = verify_provenance_receipt(&args.receipt, args.json) {
693                eprintln!("error: {error}");
694                process::exit(1);
695            }
696        }
697        Command::Completions(args) => print_completions(args.shell),
698        Command::Orchestrator(args) => {
699            if let Err(error) = commands::orchestrator::handle(args).await {
700                eprintln!("error: {error}");
701                process::exit(1);
702            }
703        }
704        Command::Playground(args) => {
705            provider_bootstrap::maybe_seed_ollama_for_playground(
706                Path::new(&args.host),
707                Path::new(&args.script),
708                args.yes,
709                args.llm.is_some(),
710                args.llm_mock.is_some(),
711            )
712            .await;
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            if let Err(error) = commands::playground::run_command(args, llm_mock_mode).await {
725                eprint!("{error}");
726                process::exit(1);
727            }
728        }
729        Command::Runs(args) => match args.command {
730            RunsCommand::Inspect(inspect) => {
731                inspect_run_record(&inspect.path, inspect.compare.as_deref())
732            }
733        },
734        Command::Replay(args) => replay_run_record(&args.path),
735        Command::Eval(args) => {
736            let llm_mock_mode = if let Some(path) = args.llm_mock.as_ref() {
737                commands::run::CliLlmMockMode::Replay {
738                    fixture_path: PathBuf::from(path),
739                }
740            } else if let Some(path) = args.llm_mock_record.as_ref() {
741                commands::run::CliLlmMockMode::Record {
742                    fixture_path: PathBuf::from(path),
743                }
744            } else {
745                commands::run::CliLlmMockMode::Off
746            };
747            eval_run_record(
748                &args.path,
749                args.compare.as_deref(),
750                args.structural_experiment.as_deref(),
751                &args.argv,
752                &llm_mock_mode,
753            )
754        }
755        Command::Repl => commands::repl::run_repl().await,
756        Command::Bench(args) => commands::bench::run_bench(&args.file, args.iterations).await,
757        Command::TestBench(args) => commands::test_bench::run(args.command).await,
758        Command::Viz(args) => commands::viz::run_viz(&args.file, args.output.as_deref()),
759        Command::Install(args) => package::install_packages(
760            args.frozen || args.locked || args.offline,
761            args.refetch.as_deref(),
762            args.offline,
763        ),
764        Command::Add(args) => package::add_package_with_registry(
765            &args.name_or_spec,
766            args.alias.as_deref(),
767            args.git.as_deref(),
768            args.tag.as_deref(),
769            args.rev.as_deref(),
770            args.branch.as_deref(),
771            args.path.as_deref(),
772            args.registry.as_deref(),
773        ),
774        Command::Update(args) => package::update_packages(args.alias.as_deref(), args.all),
775        Command::Remove(args) => package::remove_package(&args.alias),
776        Command::Lock => package::lock_packages(),
777        Command::Package(args) => match args.command {
778            PackageCommand::Search(search) => package::search_package_registry(
779                search.query.as_deref(),
780                search.registry.as_deref(),
781                search.json,
782            ),
783            PackageCommand::Info(info) => {
784                package::show_package_registry_info(&info.name, info.registry.as_deref(), info.json)
785            }
786            PackageCommand::Check(check) => {
787                package::check_package(check.package.as_deref(), check.json)
788            }
789            PackageCommand::Pack(pack) => package::pack_package(
790                pack.package.as_deref(),
791                pack.output.as_deref(),
792                pack.dry_run,
793                pack.json,
794            ),
795            PackageCommand::Docs(docs) => package::generate_package_docs(
796                docs.package.as_deref(),
797                docs.output.as_deref(),
798                docs.check,
799            ),
800            PackageCommand::Cache(cache) => match cache.command {
801                PackageCacheCommand::List => package::list_package_cache(),
802                PackageCacheCommand::Clean(clean) => package::clean_package_cache(clean.all),
803                PackageCacheCommand::Verify(verify) => {
804                    package::verify_package_cache(verify.materialized)
805                }
806            },
807        },
808        Command::Publish(args) => package::publish_package(
809            args.package.as_deref(),
810            args.dry_run,
811            args.registry.as_deref(),
812            args.json,
813        ),
814        Command::MergeCaptain(args) => match args.command {
815            MergeCaptainCommand::Run(run) => {
816                let code = commands::merge_captain::run_driver(&run);
817                if code != 0 {
818                    process::exit(code);
819                }
820            }
821            MergeCaptainCommand::Ladder(ladder) => {
822                let code = commands::merge_captain::run_ladder(&ladder);
823                if code != 0 {
824                    process::exit(code);
825                }
826            }
827            MergeCaptainCommand::Iterate(iterate) => {
828                let code = commands::merge_captain::run_iterate(&iterate);
829                if code != 0 {
830                    process::exit(code);
831                }
832            }
833            MergeCaptainCommand::Audit(audit) => {
834                let code = commands::merge_captain::run_audit(&audit);
835                if code != 0 {
836                    process::exit(code);
837                }
838            }
839            MergeCaptainCommand::Mock(mock) => {
840                let code = match mock {
841                    MergeCaptainMockCommand::Init(args) => {
842                        commands::merge_captain_mock::run_init(&args)
843                    }
844                    MergeCaptainMockCommand::Step(args) => {
845                        commands::merge_captain_mock::run_step(&args)
846                    }
847                    MergeCaptainMockCommand::Status(args) => {
848                        commands::merge_captain_mock::run_status(&args)
849                    }
850                    MergeCaptainMockCommand::Serve(args) => {
851                        commands::merge_captain_mock::run_serve(&args).await
852                    }
853                    MergeCaptainMockCommand::Cleanup(args) => {
854                        commands::merge_captain_mock::run_cleanup(&args)
855                    }
856                    MergeCaptainMockCommand::Scenarios => {
857                        commands::merge_captain_mock::run_scenarios()
858                    }
859                };
860                if code != 0 {
861                    process::exit(code);
862                }
863            }
864        },
865        Command::Persona(args) => match args.command {
866            PersonaCommand::New(new) => {
867                if let Err(error) = commands::persona_scaffold::run_new(&new) {
868                    eprintln!("error: {error}");
869                    process::exit(1);
870                }
871            }
872            PersonaCommand::Doctor(doctor) => {
873                if let Err(error) =
874                    commands::persona_doctor::run_doctor(args.manifest.as_deref(), &doctor).await
875                {
876                    eprintln!("error: {error}");
877                    process::exit(1);
878                }
879            }
880            PersonaCommand::Check(check) => {
881                commands::persona::run_check(args.manifest.as_deref(), &check)
882            }
883            PersonaCommand::List(list) => {
884                commands::persona::run_list(args.manifest.as_deref(), &list)
885            }
886            PersonaCommand::Inspect(inspect) => {
887                commands::persona::run_inspect(args.manifest.as_deref(), &inspect)
888            }
889            PersonaCommand::Status(status) => {
890                if let Err(error) = commands::persona::run_status(
891                    args.manifest.as_deref(),
892                    &args.state_dir,
893                    &status,
894                )
895                .await
896                {
897                    eprintln!("error: {error}");
898                    process::exit(1);
899                }
900            }
901            PersonaCommand::Pause(control) => {
902                if let Err(error) = commands::persona::run_pause(
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::Resume(control) => {
914                if let Err(error) = commands::persona::run_resume(
915                    args.manifest.as_deref(),
916                    &args.state_dir,
917                    &control,
918                )
919                .await
920                {
921                    eprintln!("error: {error}");
922                    process::exit(1);
923                }
924            }
925            PersonaCommand::Disable(control) => {
926                if let Err(error) = commands::persona::run_disable(
927                    args.manifest.as_deref(),
928                    &args.state_dir,
929                    &control,
930                )
931                .await
932                {
933                    eprintln!("error: {error}");
934                    process::exit(1);
935                }
936            }
937            PersonaCommand::Tick(tick) => {
938                if let Err(error) =
939                    commands::persona::run_tick(args.manifest.as_deref(), &args.state_dir, &tick)
940                        .await
941                {
942                    eprintln!("error: {error}");
943                    process::exit(1);
944                }
945            }
946            PersonaCommand::Trigger(trigger) => {
947                if let Err(error) = commands::persona::run_trigger(
948                    args.manifest.as_deref(),
949                    &args.state_dir,
950                    &trigger,
951                )
952                .await
953                {
954                    eprintln!("error: {error}");
955                    process::exit(1);
956                }
957            }
958            PersonaCommand::Spend(spend) => {
959                if let Err(error) =
960                    commands::persona::run_spend(args.manifest.as_deref(), &args.state_dir, &spend)
961                        .await
962                {
963                    eprintln!("error: {error}");
964                    process::exit(1);
965                }
966            }
967        },
968        Command::ModelInfo(args) => {
969            if !print_model_info(&args).await {
970                process::exit(1);
971            }
972        }
973        Command::ProviderCatalog(args) => print_provider_catalog(args.available_only),
974        Command::ProviderReady(args) => {
975            run_provider_ready(
976                &args.provider,
977                args.model.as_deref(),
978                args.base_url.as_deref(),
979                args.json,
980            )
981            .await
982        }
983        Command::Skills(args) => match args.command {
984            SkillsCommand::List(list) => commands::skills::run_list(&list),
985            SkillsCommand::Inspect(inspect) => commands::skills::run_inspect(&inspect),
986            SkillsCommand::Match(matcher) => commands::skills::run_match(&matcher),
987            SkillsCommand::Install(install) => commands::skills::run_install(&install),
988            SkillsCommand::New(new_args) => commands::skills::run_new(&new_args),
989        },
990        Command::DumpHighlightKeywords(args) => {
991            commands::dump_highlight_keywords::run(&args.output, args.check);
992        }
993        Command::DumpTriggerQuickref(args) => {
994            commands::dump_trigger_quickref::run(&args.output, args.check);
995        }
996        Command::DumpConnectorMatrix(args) => {
997            commands::check::connector_matrix::run_docs(&args.output, &args.sources, args.check);
998        }
999        Command::DumpProtocolArtifacts(args) => {
1000            commands::dump_protocol_artifacts::run(&args.output_dir, args.check);
1001        }
1002    }
1003}
1004
1005fn print_completions(shell: CompletionShell) {
1006    let mut command = Cli::command();
1007    let shell = clap_complete::Shell::from(shell);
1008    clap_complete::generate(shell, &mut command, "harn", &mut std::io::stdout());
1009}
1010
1011fn normalize_serve_args(mut raw_args: Vec<String>) -> Vec<String> {
1012    if raw_args.len() > 2
1013        && raw_args.get(1).is_some_and(|arg| arg == "serve")
1014        && !matches!(
1015            raw_args.get(2).map(String::as_str),
1016            Some("acp" | "a2a" | "mcp" | "-h" | "--help")
1017        )
1018    {
1019        raw_args.insert(2, "a2a".to_string());
1020    }
1021    raw_args
1022}
1023
1024fn print_version() {
1025    println!(
1026        r#"
1027 ╱▔▔╲
1028 ╱    ╲    harn v{}
1029 │ ◆  │    the agent harness language
1030 │    │
1031 ╰──╯╱
1032   ╱╱
1033"#,
1034        env!("CARGO_PKG_VERSION")
1035    );
1036}
1037
1038async fn print_model_info(args: &ModelInfoArgs) -> bool {
1039    let resolved = harn_vm::llm_config::resolve_model_info(&args.model);
1040    let api_key_result = harn_vm::llm::resolve_api_key(&resolved.provider);
1041    let api_key_set = api_key_result.is_ok();
1042    let api_key = api_key_result.unwrap_or_default();
1043    let context_window =
1044        harn_vm::llm::fetch_provider_max_context(&resolved.provider, &resolved.id, &api_key).await;
1045    let readiness = local_openai_readiness(&resolved.provider, &resolved.id, &api_key).await;
1046    let catalog = harn_vm::llm_config::model_catalog_entry(&resolved.id);
1047    let runtime_context_window = catalog
1048        .as_ref()
1049        .and_then(|entry| entry.runtime_context_window);
1050    let capabilities = harn_vm::llm::capabilities::lookup(&resolved.provider, &resolved.id);
1051    let mut payload = serde_json::json!({
1052        "alias": args.model,
1053        "id": resolved.id,
1054        "provider": resolved.provider,
1055        "resolved_alias": resolved.alias,
1056        "tool_format": resolved.tool_format,
1057        "tier": resolved.tier,
1058        "api_key_set": api_key_set,
1059        "context_window": context_window,
1060        "runtime_context_window": runtime_context_window,
1061        "readiness": readiness,
1062        "catalog": catalog,
1063        "capabilities": {
1064            "native_tools": capabilities.native_tools,
1065            "defer_loading": capabilities.defer_loading,
1066            "tool_search": capabilities.tool_search,
1067            "max_tools": capabilities.max_tools,
1068            "prompt_caching": capabilities.prompt_caching,
1069            "vision": capabilities.vision,
1070            "vision_supported": capabilities.vision_supported,
1071            "audio": capabilities.audio,
1072            "pdf": capabilities.pdf,
1073            "files_api_supported": capabilities.files_api_supported,
1074            "json_schema": capabilities.json_schema,
1075            "thinking": !capabilities.thinking_modes.is_empty(),
1076            "thinking_modes": capabilities.thinking_modes,
1077            "interleaved_thinking_supported": capabilities.interleaved_thinking_supported,
1078            "anthropic_beta_features": capabilities.anthropic_beta_features,
1079            "preserve_thinking": capabilities.preserve_thinking,
1080            "server_parser": capabilities.server_parser,
1081            "honors_chat_template_kwargs": capabilities.honors_chat_template_kwargs,
1082            "recommended_endpoint": capabilities.recommended_endpoint,
1083            "text_tool_wire_format_supported": capabilities.text_tool_wire_format_supported,
1084        },
1085        "qc_default_model": harn_vm::llm_config::qc_default_model(&resolved.provider),
1086    });
1087
1088    let should_verify = args.verify || args.warm;
1089    let mut ok = true;
1090    if should_verify {
1091        if resolved.provider == "ollama" {
1092            let mut readiness = harn_vm::llm::OllamaReadinessOptions::new(resolved.id.clone());
1093            readiness.warm = args.warm;
1094            readiness.keep_alive = args
1095                .keep_alive
1096                .as_deref()
1097                .and_then(harn_vm::llm::normalize_ollama_keep_alive);
1098            let result = harn_vm::llm::ollama_readiness(readiness).await;
1099            ok = result.valid;
1100            payload["readiness"] = serde_json::to_value(&result).unwrap_or_else(|error| {
1101                serde_json::json!({
1102                    "valid": false,
1103                    "status": "serialization_error",
1104                    "message": format!("failed to serialize readiness result: {error}"),
1105                })
1106            });
1107        } else {
1108            ok = false;
1109            payload["readiness"] = serde_json::json!({
1110                "valid": false,
1111                "status": "unsupported_provider",
1112                "message": format!(
1113                    "model-info --verify is only supported for Ollama models; resolved provider is '{}'",
1114                    resolved.provider
1115                ),
1116                "provider": resolved.provider,
1117            });
1118        }
1119    }
1120
1121    println!(
1122        "{}",
1123        serde_json::to_string(&payload).unwrap_or_else(|error| {
1124            command_error(&format!("failed to serialize model info: {error}"))
1125        })
1126    );
1127    ok
1128}
1129
1130async fn local_openai_readiness(
1131    provider: &str,
1132    model: &str,
1133    api_key: &str,
1134) -> Option<serde_json::Value> {
1135    let def = harn_vm::llm_config::provider_config(provider)?;
1136    if def.auth_style != "none" || !harn_vm::llm::supports_model_readiness_probe(&def) {
1137        return None;
1138    }
1139    let readiness = harn_vm::llm::probe_openai_compatible_model(provider, model, api_key).await;
1140    Some(serde_json::json!({
1141        "valid": readiness.valid,
1142        "category": readiness.category,
1143        "message": readiness.message,
1144        "provider": readiness.provider,
1145        "model": readiness.model,
1146        "url": readiness.url,
1147        "status": readiness.status,
1148        "available_models": readiness.available_models,
1149    }))
1150}
1151
1152fn print_provider_catalog(available_only: bool) {
1153    let provider_names = if available_only {
1154        harn_vm::llm_config::available_provider_names()
1155    } else {
1156        harn_vm::llm_config::provider_names()
1157    };
1158    let providers: Vec<_> = provider_names
1159        .into_iter()
1160        .filter_map(|name| {
1161            harn_vm::llm_config::provider_config(&name).map(|def| {
1162                serde_json::json!({
1163                    "name": name,
1164                    "display_name": def.display_name,
1165                    "icon": def.icon,
1166                    "base_url": harn_vm::llm_config::resolve_base_url(&def),
1167                    "base_url_env": def.base_url_env,
1168                    "auth_style": def.auth_style,
1169                    "auth_envs": harn_vm::llm_config::auth_env_names(&def.auth_env),
1170                    "auth_available": harn_vm::llm_config::provider_key_available(&name),
1171                    "features": def.features,
1172                    "cost_per_1k_in": def.cost_per_1k_in,
1173                    "cost_per_1k_out": def.cost_per_1k_out,
1174                    "latency_p50_ms": def.latency_p50_ms,
1175                })
1176            })
1177        })
1178        .collect();
1179    let models: Vec<_> = harn_vm::llm_config::model_catalog_entries()
1180        .into_iter()
1181        .map(|(id, model)| {
1182            serde_json::json!({
1183                "id": id,
1184                "name": model.name,
1185                "provider": model.provider,
1186                "context_window": model.context_window,
1187                "runtime_context_window": model.runtime_context_window,
1188                "stream_timeout": model.stream_timeout,
1189                "capabilities": model.capabilities,
1190                "pricing": model.pricing,
1191            })
1192        })
1193        .collect();
1194    let aliases: Vec<_> = harn_vm::llm_config::alias_entries()
1195        .into_iter()
1196        .map(|(name, alias)| {
1197            serde_json::json!({
1198                "name": name,
1199                "id": alias.id,
1200                "provider": alias.provider,
1201                "tool_format": alias.tool_format,
1202            })
1203        })
1204        .collect();
1205    let payload = serde_json::json!({
1206        "providers": providers,
1207        "known_model_names": harn_vm::llm_config::known_model_names(),
1208        "available_providers": harn_vm::llm_config::available_provider_names(),
1209        "aliases": aliases,
1210        "models": models,
1211        "qc_defaults": harn_vm::llm_config::qc_defaults(),
1212    });
1213    println!(
1214        "{}",
1215        serde_json::to_string(&payload).unwrap_or_else(|error| {
1216            command_error(&format!("failed to serialize provider catalog: {error}"))
1217        })
1218    );
1219}
1220
1221async fn run_provider_ready(
1222    provider: &str,
1223    model: Option<&str>,
1224    base_url: Option<&str>,
1225    json: bool,
1226) {
1227    let readiness =
1228        harn_vm::llm::readiness::probe_provider_readiness(provider, model, base_url).await;
1229    if json {
1230        match serde_json::to_string_pretty(&readiness) {
1231            Ok(payload) => println!("{payload}"),
1232            Err(error) => command_error(&format!("failed to serialize readiness result: {error}")),
1233        }
1234    } else if readiness.ok {
1235        println!("{}", readiness.message);
1236    } else {
1237        eprintln!("{}", readiness.message);
1238    }
1239    if !readiness.ok {
1240        process::exit(1);
1241    }
1242}
1243
1244fn command_error(message: &str) -> ! {
1245    Cli::command()
1246        .error(ErrorKind::ValueValidation, message)
1247        .exit()
1248}
1249
1250fn verify_provenance_receipt(path: &str, json: bool) -> Result<(), String> {
1251    let raw =
1252        fs::read_to_string(path).map_err(|error| format!("failed to read {path}: {error}"))?;
1253    let receipt: harn_vm::ProvenanceReceipt = serde_json::from_str(&raw)
1254        .map_err(|error| format!("failed to parse provenance receipt {path}: {error}"))?;
1255    let report = harn_vm::verify_receipt(&receipt);
1256    if json {
1257        println!(
1258            "{}",
1259            serde_json::to_string_pretty(&report).map_err(|error| error.to_string())?
1260        );
1261    } else if report.verified {
1262        println!(
1263            "verified receipt={} events={} receipt_hash={} event_root_hash={}",
1264            report.receipt_id.unwrap_or_else(|| "-".to_string()),
1265            report.event_count,
1266            report.receipt_hash.unwrap_or_else(|| "-".to_string()),
1267            report.event_root_hash.unwrap_or_else(|| "-".to_string())
1268        );
1269    } else {
1270        println!(
1271            "failed receipt={} events={}",
1272            report.receipt_id.unwrap_or_else(|| "-".to_string()),
1273            report.event_count
1274        );
1275        for error in &report.errors {
1276            println!("  {error}");
1277        }
1278        return Err("provenance receipt verification failed".to_string());
1279    }
1280    Ok(())
1281}
1282
1283fn load_run_record_or_exit(path: &Path) -> harn_vm::orchestration::RunRecord {
1284    match harn_vm::orchestration::load_run_record(path) {
1285        Ok(run) => run,
1286        Err(error) => {
1287            eprintln!("Failed to load run record: {error}");
1288            process::exit(1);
1289        }
1290    }
1291}
1292
1293fn load_eval_suite_manifest_or_exit(path: &Path) -> harn_vm::orchestration::EvalSuiteManifest {
1294    harn_vm::orchestration::load_eval_suite_manifest(path).unwrap_or_else(|error| {
1295        eprintln!("Failed to load eval manifest {}: {error}", path.display());
1296        process::exit(1);
1297    })
1298}
1299
1300fn load_eval_pack_manifest_or_exit(path: &Path) -> harn_vm::orchestration::EvalPackManifest {
1301    harn_vm::orchestration::load_eval_pack_manifest(path).unwrap_or_else(|error| {
1302        eprintln!("Failed to load eval pack {}: {error}", path.display());
1303        process::exit(1);
1304    })
1305}
1306
1307fn load_persona_eval_ladder_manifest_or_exit(
1308    path: &Path,
1309) -> harn_vm::orchestration::PersonaEvalLadderManifest {
1310    harn_vm::orchestration::load_persona_eval_ladder_manifest(path).unwrap_or_else(|error| {
1311        eprintln!(
1312            "Failed to load persona eval ladder {}: {error}",
1313            path.display()
1314        );
1315        process::exit(1);
1316    })
1317}
1318
1319fn file_looks_like_eval_manifest(path: &Path) -> bool {
1320    if path.file_name().and_then(|name| name.to_str()) == Some("harn.eval.toml") {
1321        return true;
1322    }
1323    if path.extension().and_then(|ext| ext.to_str()) == Some("toml") {
1324        let Ok(content) = fs::read_to_string(path) else {
1325            return false;
1326        };
1327        return toml::from_str::<harn_vm::orchestration::EvalPackManifest>(&content)
1328            .is_ok_and(|manifest| !manifest.cases.is_empty() || !manifest.ladders.is_empty());
1329    }
1330    let Ok(content) = fs::read_to_string(path) else {
1331        return false;
1332    };
1333    let Ok(json) = serde_json::from_str::<serde_json::Value>(&content) else {
1334        return false;
1335    };
1336    json.get("_type").and_then(|value| value.as_str()) == Some("eval_suite_manifest")
1337        || json.get("cases").is_some()
1338}
1339
1340fn file_looks_like_eval_pack_manifest(path: &Path) -> bool {
1341    if path.file_name().and_then(|name| name.to_str()) == Some("harn.eval.toml") {
1342        return true;
1343    }
1344    if path.extension().and_then(|ext| ext.to_str()) == Some("toml") {
1345        return file_looks_like_eval_manifest(path);
1346    }
1347    let Ok(content) = fs::read_to_string(path) else {
1348        return false;
1349    };
1350    let Ok(json) = serde_json::from_str::<serde_json::Value>(&content) else {
1351        return false;
1352    };
1353    json.get("version").is_some()
1354        && (json.get("cases").is_some() || json.get("ladders").is_some())
1355        && json.get("_type").and_then(|value| value.as_str()) != Some("eval_suite_manifest")
1356}
1357
1358fn file_looks_like_persona_eval_ladder_manifest(path: &Path) -> bool {
1359    let Ok(content) = fs::read_to_string(path) else {
1360        return false;
1361    };
1362    if path.extension().and_then(|ext| ext.to_str()) == Some("json") {
1363        let Ok(json) = serde_json::from_str::<serde_json::Value>(&content) else {
1364            return false;
1365        };
1366        return json.get("_type").and_then(|value| value.as_str())
1367            == Some("persona_eval_ladder_manifest")
1368            || json.get("timeout_tiers").is_some()
1369            || json.get("timeout-tiers").is_some();
1370    }
1371    toml::from_str::<harn_vm::orchestration::PersonaEvalLadderManifest>(&content).is_ok_and(
1372        |manifest| {
1373            manifest
1374                .type_name
1375                .eq_ignore_ascii_case("persona_eval_ladder_manifest")
1376                || (!manifest.timeout_tiers.is_empty() && manifest.backend.path.is_some())
1377        },
1378    )
1379}
1380
1381fn collect_run_record_paths(path: &str) -> Vec<PathBuf> {
1382    let path = Path::new(path);
1383    if path.is_file() {
1384        return vec![path.to_path_buf()];
1385    }
1386    if path.is_dir() {
1387        let mut entries: Vec<PathBuf> = fs::read_dir(path)
1388            .unwrap_or_else(|error| {
1389                eprintln!("Failed to read run directory {}: {error}", path.display());
1390                process::exit(1);
1391            })
1392            .filter_map(|entry| entry.ok().map(|entry| entry.path()))
1393            .filter(|entry| entry.extension().and_then(|ext| ext.to_str()) == Some("json"))
1394            .collect();
1395        entries.sort();
1396        return entries;
1397    }
1398    eprintln!("Run path does not exist: {}", path.display());
1399    process::exit(1);
1400}
1401
1402fn print_run_diff(diff: &harn_vm::orchestration::RunDiffReport) {
1403    println!(
1404        "Diff: {} -> {} [{} -> {}]",
1405        diff.left_run_id, diff.right_run_id, diff.left_status, diff.right_status
1406    );
1407    println!("Identical: {}", diff.identical);
1408    println!("Stage diffs: {}", diff.stage_diffs.len());
1409    println!("Tool diffs: {}", diff.tool_diffs.len());
1410    println!("Observability diffs: {}", diff.observability_diffs.len());
1411    println!("Transition delta: {}", diff.transition_count_delta);
1412    println!("Artifact delta: {}", diff.artifact_count_delta);
1413    println!("Checkpoint delta: {}", diff.checkpoint_count_delta);
1414    for stage in &diff.stage_diffs {
1415        println!("- {} [{}]", stage.node_id, stage.change);
1416        for detail in &stage.details {
1417            println!("  {}", detail);
1418        }
1419    }
1420    for tool in &diff.tool_diffs {
1421        println!("- tool {} [{}]", tool.tool_name, tool.args_hash);
1422        println!("  left: {:?}", tool.left_result);
1423        println!("  right: {:?}", tool.right_result);
1424    }
1425    for item in &diff.observability_diffs {
1426        println!("- {} [{}]", item.label, item.section);
1427        for detail in &item.details {
1428            println!("  {}", detail);
1429        }
1430    }
1431}
1432
1433fn inspect_run_record(path: &str, compare: Option<&str>) {
1434    let run = load_run_record_or_exit(Path::new(path));
1435    println!("Run: {}", run.id);
1436    println!(
1437        "Workflow: {}",
1438        run.workflow_name
1439            .clone()
1440            .unwrap_or_else(|| run.workflow_id.clone())
1441    );
1442    println!("Status: {}", run.status);
1443    println!("Task: {}", run.task);
1444    println!("Stages: {}", run.stages.len());
1445    println!("Artifacts: {}", run.artifacts.len());
1446    println!("Transitions: {}", run.transitions.len());
1447    println!("Checkpoints: {}", run.checkpoints.len());
1448    println!("HITL questions: {}", run.hitl_questions.len());
1449    if let Some(observability) = &run.observability {
1450        println!("Planner rounds: {}", observability.planner_rounds.len());
1451        println!("Research facts: {}", observability.research_fact_count);
1452        println!("Workers: {}", observability.worker_lineage.len());
1453        println!(
1454            "Action graph: {} nodes / {} edges",
1455            observability.action_graph_nodes.len(),
1456            observability.action_graph_edges.len()
1457        );
1458        println!(
1459            "Transcript pointers: {}",
1460            observability.transcript_pointers.len()
1461        );
1462        println!("Daemon events: {}", observability.daemon_events.len());
1463    }
1464    if let Some(parent_worker_id) = run
1465        .metadata
1466        .get("parent_worker_id")
1467        .and_then(|value| value.as_str())
1468    {
1469        println!("Parent worker: {}", parent_worker_id);
1470    }
1471    if let Some(parent_stage_id) = run
1472        .metadata
1473        .get("parent_stage_id")
1474        .and_then(|value| value.as_str())
1475    {
1476        println!("Parent stage: {}", parent_stage_id);
1477    }
1478    if run
1479        .metadata
1480        .get("delegated")
1481        .and_then(|value| value.as_bool())
1482        .unwrap_or(false)
1483    {
1484        println!("Delegated: true");
1485    }
1486    println!(
1487        "Pending nodes: {}",
1488        if run.pending_nodes.is_empty() {
1489            "-".to_string()
1490        } else {
1491            run.pending_nodes.join(", ")
1492        }
1493    );
1494    println!(
1495        "Replay fixture: {}",
1496        if run.replay_fixture.is_some() {
1497            "embedded"
1498        } else {
1499            "derived"
1500        }
1501    );
1502    for stage in &run.stages {
1503        let worker = stage.metadata.get("worker");
1504        let worker_suffix = worker
1505            .and_then(|value| value.get("name"))
1506            .and_then(|value| value.as_str())
1507            .map(|name| format!(" worker={name}"))
1508            .unwrap_or_default();
1509        println!(
1510            "- {} [{}] status={} outcome={} branch={}{}",
1511            stage.node_id,
1512            stage.kind,
1513            stage.status,
1514            stage.outcome,
1515            stage.branch.clone().unwrap_or_else(|| "-".to_string()),
1516            worker_suffix,
1517        );
1518        if let Some(worker) = worker {
1519            if let Some(worker_id) = worker.get("id").and_then(|value| value.as_str()) {
1520                println!("  worker_id: {}", worker_id);
1521            }
1522            if let Some(child_run_id) = worker.get("child_run_id").and_then(|value| value.as_str())
1523            {
1524                println!("  child_run_id: {}", child_run_id);
1525            }
1526            if let Some(child_run_path) = worker
1527                .get("child_run_path")
1528                .and_then(|value| value.as_str())
1529            {
1530                println!("  child_run_path: {}", child_run_path);
1531            }
1532        }
1533    }
1534    if let Some(observability) = &run.observability {
1535        for round in &observability.planner_rounds {
1536            println!(
1537                "- planner {} iterations={} llm_calls={} tools={} research_facts={}",
1538                round.node_id,
1539                round.iteration_count,
1540                round.llm_call_count,
1541                round.tool_execution_count,
1542                round.research_facts.len()
1543            );
1544        }
1545        for pointer in &observability.transcript_pointers {
1546            println!(
1547                "- transcript {} [{}] available={} {}",
1548                pointer.label,
1549                pointer.kind,
1550                pointer.available,
1551                pointer
1552                    .path
1553                    .clone()
1554                    .unwrap_or_else(|| pointer.location.clone())
1555            );
1556        }
1557        for event in &observability.daemon_events {
1558            println!(
1559                "- daemon {} [{:?}] at {}",
1560                event.name, event.kind, event.timestamp
1561            );
1562            println!("  id: {}", event.daemon_id);
1563            println!("  persist_path: {}", event.persist_path);
1564            if let Some(summary) = &event.payload_summary {
1565                println!("  payload: {}", summary);
1566            }
1567        }
1568    }
1569    if let Some(compare_path) = compare {
1570        let baseline = load_run_record_or_exit(Path::new(compare_path));
1571        print_run_diff(&harn_vm::orchestration::diff_run_records(&baseline, &run));
1572    }
1573}
1574
1575fn replay_run_record(path: &str) {
1576    let run = load_run_record_or_exit(Path::new(path));
1577    println!("Replay: {}", run.id);
1578    for stage in &run.stages {
1579        println!(
1580            "[{}] status={} outcome={} branch={}",
1581            stage.node_id,
1582            stage.status,
1583            stage.outcome,
1584            stage.branch.clone().unwrap_or_else(|| "-".to_string())
1585        );
1586        if let Some(text) = &stage.visible_text {
1587            println!("  visible: {}", text);
1588        }
1589        if let Some(verification) = &stage.verification {
1590            println!("  verification: {}", verification);
1591        }
1592    }
1593    if let Some(transcript) = &run.transcript {
1594        println!(
1595            "Transcript events persisted: {}",
1596            transcript["events"]
1597                .as_array()
1598                .map(|v| v.len())
1599                .unwrap_or(0)
1600        );
1601    }
1602    let fixture = run
1603        .replay_fixture
1604        .clone()
1605        .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(&run));
1606    let report = harn_vm::orchestration::evaluate_run_against_fixture(&run, &fixture);
1607    println!(
1608        "Embedded replay fixture: {}",
1609        if report.pass { "PASS" } else { "FAIL" }
1610    );
1611    for transition in &run.transitions {
1612        println!(
1613            "transition {} -> {} ({})",
1614            transition
1615                .from_node_id
1616                .clone()
1617                .unwrap_or_else(|| "start".to_string()),
1618            transition.to_node_id,
1619            transition
1620                .branch
1621                .clone()
1622                .unwrap_or_else(|| "default".to_string())
1623        );
1624    }
1625}
1626
1627fn eval_run_record(
1628    path: &str,
1629    compare: Option<&str>,
1630    structural_experiment: Option<&str>,
1631    argv: &[String],
1632    llm_mock_mode: &commands::run::CliLlmMockMode,
1633) {
1634    if let Some(experiment) = structural_experiment {
1635        let path_buf = PathBuf::from(path);
1636        if !path_buf.is_file() || path_buf.extension().and_then(|ext| ext.to_str()) != Some("harn")
1637        {
1638            eprintln!(
1639                "--structural-experiment currently requires a .harn pipeline path, got {}",
1640                path
1641            );
1642            process::exit(1);
1643        }
1644        if compare.is_some() {
1645            eprintln!("--compare cannot be combined with --structural-experiment");
1646            process::exit(1);
1647        }
1648        if matches!(llm_mock_mode, commands::run::CliLlmMockMode::Record { .. }) {
1649            eprintln!("--llm-mock-record cannot be combined with --structural-experiment");
1650            process::exit(1);
1651        }
1652        let path_buf = fs::canonicalize(&path_buf).unwrap_or_else(|error| {
1653            command_error(&format!(
1654                "failed to canonicalize structural eval pipeline {}: {error}",
1655                path_buf.display()
1656            ))
1657        });
1658        run_structural_experiment_eval(&path_buf, experiment, argv, llm_mock_mode);
1659        return;
1660    }
1661
1662    let path_buf = PathBuf::from(path);
1663    if path_buf.is_file() && file_looks_like_persona_eval_ladder_manifest(&path_buf) {
1664        if compare.is_some() {
1665            eprintln!("--compare is not supported with persona eval ladder manifests");
1666            process::exit(1);
1667        }
1668        let manifest = load_persona_eval_ladder_manifest_or_exit(&path_buf);
1669        let report =
1670            harn_vm::orchestration::run_persona_eval_ladder(&manifest).unwrap_or_else(|error| {
1671                eprintln!(
1672                    "Failed to evaluate persona eval ladder {}: {error}",
1673                    path_buf.display()
1674                );
1675                process::exit(1);
1676            });
1677        print_persona_ladder_report(&report);
1678        if !report.pass {
1679            process::exit(1);
1680        }
1681        return;
1682    }
1683
1684    if path_buf.is_file() && file_looks_like_eval_pack_manifest(&path_buf) {
1685        if compare.is_some() {
1686            eprintln!("--compare is not supported with eval pack manifests");
1687            process::exit(1);
1688        }
1689        let manifest = load_eval_pack_manifest_or_exit(&path_buf);
1690        let report = harn_vm::orchestration::evaluate_eval_pack_manifest(&manifest).unwrap_or_else(
1691            |error| {
1692                eprintln!(
1693                    "Failed to evaluate eval pack {}: {error}",
1694                    path_buf.display()
1695                );
1696                process::exit(1);
1697            },
1698        );
1699        print_eval_pack_report(&report);
1700        if !report.pass {
1701            process::exit(1);
1702        }
1703        return;
1704    }
1705
1706    if path_buf.is_file() && file_looks_like_eval_manifest(&path_buf) {
1707        if compare.is_some() {
1708            eprintln!("--compare is not supported with eval suite manifests");
1709            process::exit(1);
1710        }
1711        let manifest = load_eval_suite_manifest_or_exit(&path_buf);
1712        let suite = harn_vm::orchestration::evaluate_run_suite_manifest(&manifest).unwrap_or_else(
1713            |error| {
1714                eprintln!(
1715                    "Failed to evaluate manifest {}: {error}",
1716                    path_buf.display()
1717                );
1718                process::exit(1);
1719            },
1720        );
1721        println!(
1722            "{} {} passed, {} failed, {} total",
1723            if suite.pass { "PASS" } else { "FAIL" },
1724            suite.passed,
1725            suite.failed,
1726            suite.total
1727        );
1728        for case in &suite.cases {
1729            println!(
1730                "- {} [{}] {}",
1731                case.label.clone().unwrap_or_else(|| case.run_id.clone()),
1732                case.workflow_id,
1733                if case.pass { "PASS" } else { "FAIL" }
1734            );
1735            if let Some(path) = &case.source_path {
1736                println!("  path: {}", path);
1737            }
1738            if let Some(comparison) = &case.comparison {
1739                println!("  baseline identical: {}", comparison.identical);
1740                if !comparison.identical {
1741                    println!(
1742                        "  baseline status: {} -> {}",
1743                        comparison.left_status, comparison.right_status
1744                    );
1745                }
1746            }
1747            for failure in &case.failures {
1748                println!("  {}", failure);
1749            }
1750        }
1751        if !suite.pass {
1752            process::exit(1);
1753        }
1754        return;
1755    }
1756
1757    let paths = collect_run_record_paths(path);
1758    if paths.len() > 1 {
1759        let mut cases = Vec::new();
1760        for path in &paths {
1761            let run = load_run_record_or_exit(path);
1762            let fixture = run
1763                .replay_fixture
1764                .clone()
1765                .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(&run));
1766            cases.push((run, fixture, Some(path.display().to_string())));
1767        }
1768        let suite = harn_vm::orchestration::evaluate_run_suite(cases);
1769        println!(
1770            "{} {} passed, {} failed, {} total",
1771            if suite.pass { "PASS" } else { "FAIL" },
1772            suite.passed,
1773            suite.failed,
1774            suite.total
1775        );
1776        for case in &suite.cases {
1777            println!(
1778                "- {} [{}] {}",
1779                case.run_id,
1780                case.workflow_id,
1781                if case.pass { "PASS" } else { "FAIL" }
1782            );
1783            if let Some(path) = &case.source_path {
1784                println!("  path: {}", path);
1785            }
1786            if let Some(comparison) = &case.comparison {
1787                println!("  baseline identical: {}", comparison.identical);
1788            }
1789            for failure in &case.failures {
1790                println!("  {}", failure);
1791            }
1792        }
1793        if !suite.pass {
1794            process::exit(1);
1795        }
1796        return;
1797    }
1798
1799    let run = load_run_record_or_exit(&paths[0]);
1800    let fixture = run
1801        .replay_fixture
1802        .clone()
1803        .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(&run));
1804    let report = harn_vm::orchestration::evaluate_run_against_fixture(&run, &fixture);
1805    println!("{}", if report.pass { "PASS" } else { "FAIL" });
1806    println!("Stages: {}", report.stage_count);
1807    if let Some(compare_path) = compare {
1808        let baseline = load_run_record_or_exit(Path::new(compare_path));
1809        print_run_diff(&harn_vm::orchestration::diff_run_records(&baseline, &run));
1810    }
1811    if !report.failures.is_empty() {
1812        for failure in &report.failures {
1813            println!("- {}", failure);
1814        }
1815    }
1816    if !report.pass {
1817        process::exit(1);
1818    }
1819}
1820
1821fn print_eval_pack_report(report: &harn_vm::orchestration::EvalPackReport) {
1822    println!(
1823        "{} {} passed, {} blocking failed, {} warning, {} informational, {} total",
1824        if report.pass { "PASS" } else { "FAIL" },
1825        report.passed,
1826        report.blocking_failed,
1827        report.warning_failed,
1828        report.informational_failed,
1829        report.total
1830    );
1831    for case in &report.cases {
1832        println!(
1833            "- {} [{}] {} ({})",
1834            case.label,
1835            case.workflow_id,
1836            if case.pass { "PASS" } else { "FAIL" },
1837            case.severity
1838        );
1839        if let Some(path) = &case.source_path {
1840            println!("  path: {}", path);
1841        }
1842        if let Some(comparison) = &case.comparison {
1843            println!("  baseline identical: {}", comparison.identical);
1844            if !comparison.identical {
1845                println!(
1846                    "  baseline status: {} -> {}",
1847                    comparison.left_status, comparison.right_status
1848                );
1849            }
1850        }
1851        for failure in &case.failures {
1852            println!("  {}", failure);
1853        }
1854        for warning in &case.warnings {
1855            println!("  warning: {}", warning);
1856        }
1857        for item in &case.informational {
1858            println!("  info: {}", item);
1859        }
1860    }
1861    for ladder in &report.ladders {
1862        println!(
1863            "- ladder {} [{}] {} ({}) first_correct={}/{}",
1864            ladder.id,
1865            ladder.persona,
1866            if ladder.pass { "PASS" } else { "FAIL" },
1867            ladder.severity,
1868            ladder.first_correct_route.as_deref().unwrap_or("<none>"),
1869            ladder.first_correct_tier.as_deref().unwrap_or("<none>")
1870        );
1871        println!("  artifacts: {}", ladder.artifact_root);
1872        for tier in &ladder.tiers {
1873            println!(
1874                "  - {} [{}] {} tools={} models={} latency={}ms cost=${:.6}",
1875                tier.timeout_tier,
1876                tier.route_id,
1877                tier.outcome,
1878                tier.tool_calls,
1879                tier.model_calls,
1880                tier.latency_ms,
1881                tier.cost_usd
1882            );
1883            for reason in &tier.degradation_reasons {
1884                println!("    {}", reason);
1885            }
1886        }
1887    }
1888}
1889
1890fn print_persona_ladder_report(report: &harn_vm::orchestration::PersonaEvalLadderReport) {
1891    println!(
1892        "{} ladder {} passed, {} degraded/looped, {} total",
1893        if report.pass { "PASS" } else { "FAIL" },
1894        report.passed,
1895        report.failed,
1896        report.total
1897    );
1898    println!(
1899        "first_correct: {}/{}",
1900        report.first_correct_route.as_deref().unwrap_or("<none>"),
1901        report.first_correct_tier.as_deref().unwrap_or("<none>")
1902    );
1903    println!("artifacts: {}", report.artifact_root);
1904    for tier in &report.tiers {
1905        println!(
1906            "- {} [{}] {} tools={} models={} latency={}ms cost=${:.6}",
1907            tier.timeout_tier,
1908            tier.route_id,
1909            tier.outcome,
1910            tier.tool_calls,
1911            tier.model_calls,
1912            tier.latency_ms,
1913            tier.cost_usd
1914        );
1915        for reason in &tier.degradation_reasons {
1916            println!("  {}", reason);
1917        }
1918    }
1919}
1920
1921fn run_package_evals() {
1922    let paths = package::load_package_eval_pack_paths(None).unwrap_or_else(|error| {
1923        eprintln!("{error}");
1924        process::exit(1);
1925    });
1926    let mut all_pass = true;
1927    for path in &paths {
1928        println!("Eval pack: {}", path.display());
1929        let manifest = load_eval_pack_manifest_or_exit(path);
1930        let report = harn_vm::orchestration::evaluate_eval_pack_manifest(&manifest).unwrap_or_else(
1931            |error| {
1932                eprintln!("Failed to evaluate eval pack {}: {error}", path.display());
1933                process::exit(1);
1934            },
1935        );
1936        print_eval_pack_report(&report);
1937        all_pass &= report.pass;
1938    }
1939    if !all_pass {
1940        process::exit(1);
1941    }
1942}
1943
1944fn run_structural_experiment_eval(
1945    path: &Path,
1946    experiment: &str,
1947    argv: &[String],
1948    llm_mock_mode: &commands::run::CliLlmMockMode,
1949) {
1950    let baseline_dir = tempfile::Builder::new()
1951        .prefix("harn-eval-baseline-")
1952        .tempdir()
1953        .unwrap_or_else(|error| {
1954            command_error(&format!("failed to create baseline tempdir: {error}"))
1955        });
1956    let variant_dir = tempfile::Builder::new()
1957        .prefix("harn-eval-variant-")
1958        .tempdir()
1959        .unwrap_or_else(|error| {
1960            command_error(&format!("failed to create variant tempdir: {error}"))
1961        });
1962
1963    let baseline = spawn_eval_pipeline_run(path, baseline_dir.path(), None, argv, llm_mock_mode);
1964    if !baseline.status.success() {
1965        relay_subprocess_failure("baseline", &baseline);
1966    }
1967
1968    let variant = spawn_eval_pipeline_run(
1969        path,
1970        variant_dir.path(),
1971        Some(experiment),
1972        argv,
1973        llm_mock_mode,
1974    );
1975    if !variant.status.success() {
1976        relay_subprocess_failure("variant", &variant);
1977    }
1978
1979    let baseline_runs = collect_structural_eval_runs(baseline_dir.path());
1980    let variant_runs = collect_structural_eval_runs(variant_dir.path());
1981    if baseline_runs.is_empty() || variant_runs.is_empty() {
1982        eprintln!(
1983            "structural eval expected workflow run records under {} and {}, but one side was empty",
1984            baseline_dir.path().display(),
1985            variant_dir.path().display()
1986        );
1987        process::exit(1);
1988    }
1989    if baseline_runs.len() != variant_runs.len() {
1990        eprintln!(
1991            "structural eval produced different run counts: baseline={} variant={}",
1992            baseline_runs.len(),
1993            variant_runs.len()
1994        );
1995        process::exit(1);
1996    }
1997
1998    let mut baseline_ok = 0usize;
1999    let mut variant_ok = 0usize;
2000    let mut any_failures = false;
2001
2002    println!("Structural experiment: {}", experiment);
2003    println!("Cases: {}", baseline_runs.len());
2004    for (baseline_run, variant_run) in baseline_runs.iter().zip(variant_runs.iter()) {
2005        let baseline_fixture = baseline_run
2006            .replay_fixture
2007            .clone()
2008            .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(baseline_run));
2009        let variant_fixture = variant_run
2010            .replay_fixture
2011            .clone()
2012            .unwrap_or_else(|| harn_vm::orchestration::replay_fixture_from_run(variant_run));
2013        let baseline_report =
2014            harn_vm::orchestration::evaluate_run_against_fixture(baseline_run, &baseline_fixture);
2015        let variant_report =
2016            harn_vm::orchestration::evaluate_run_against_fixture(variant_run, &variant_fixture);
2017        let diff = harn_vm::orchestration::diff_run_records(baseline_run, variant_run);
2018        if baseline_report.pass {
2019            baseline_ok += 1;
2020        }
2021        if variant_report.pass {
2022            variant_ok += 1;
2023        }
2024        any_failures |= !baseline_report.pass || !variant_report.pass;
2025        println!(
2026            "- {} [{}]",
2027            variant_run
2028                .workflow_name
2029                .clone()
2030                .unwrap_or_else(|| variant_run.workflow_id.clone()),
2031            variant_run.task
2032        );
2033        println!(
2034            "  baseline: {}",
2035            if baseline_report.pass { "PASS" } else { "FAIL" }
2036        );
2037        for failure in &baseline_report.failures {
2038            println!("    {}", failure);
2039        }
2040        println!(
2041            "  variant: {}",
2042            if variant_report.pass { "PASS" } else { "FAIL" }
2043        );
2044        for failure in &variant_report.failures {
2045            println!("    {}", failure);
2046        }
2047        println!("  diff identical: {}", diff.identical);
2048        println!("  stage diffs: {}", diff.stage_diffs.len());
2049        println!("  tool diffs: {}", diff.tool_diffs.len());
2050        println!("  observability diffs: {}", diff.observability_diffs.len());
2051    }
2052
2053    println!("Baseline {} / {} passed", baseline_ok, baseline_runs.len());
2054    println!("Variant {} / {} passed", variant_ok, variant_runs.len());
2055
2056    if any_failures {
2057        process::exit(1);
2058    }
2059}
2060
2061fn spawn_eval_pipeline_run(
2062    path: &Path,
2063    run_dir: &Path,
2064    structural_experiment: Option<&str>,
2065    argv: &[String],
2066    llm_mock_mode: &commands::run::CliLlmMockMode,
2067) -> std::process::Output {
2068    let exe = env::current_exe().unwrap_or_else(|error| {
2069        command_error(&format!("failed to resolve current executable: {error}"))
2070    });
2071    let mut command = std::process::Command::new(exe);
2072    command.current_dir(path.parent().unwrap_or_else(|| Path::new(".")));
2073    command.arg("run");
2074    match llm_mock_mode {
2075        commands::run::CliLlmMockMode::Off => {}
2076        commands::run::CliLlmMockMode::Replay { fixture_path } => {
2077            command
2078                .arg("--llm-mock")
2079                .arg(absolute_cli_path(fixture_path));
2080        }
2081        commands::run::CliLlmMockMode::Record { fixture_path } => {
2082            command
2083                .arg("--llm-mock-record")
2084                .arg(absolute_cli_path(fixture_path));
2085        }
2086    }
2087    command.arg(path);
2088    if !argv.is_empty() {
2089        command.arg("--");
2090        command.args(argv);
2091    }
2092    command.env(harn_vm::runtime_paths::HARN_RUN_DIR_ENV, run_dir);
2093    if let Some(experiment) = structural_experiment {
2094        command.env("HARN_STRUCTURAL_EXPERIMENT", experiment);
2095    }
2096    command.output().unwrap_or_else(|error| {
2097        command_error(&format!(
2098            "failed to spawn `harn run {}` for structural eval: {error}",
2099            path.display()
2100        ))
2101    })
2102}
2103
2104fn absolute_cli_path(path: &Path) -> PathBuf {
2105    if path.is_absolute() {
2106        return path.to_path_buf();
2107    }
2108    env::current_dir()
2109        .unwrap_or_else(|_| PathBuf::from("."))
2110        .join(path)
2111}
2112
2113fn relay_subprocess_failure(label: &str, output: &std::process::Output) -> ! {
2114    let stdout = String::from_utf8_lossy(&output.stdout);
2115    let stderr = String::from_utf8_lossy(&output.stderr);
2116    if !stdout.trim().is_empty() {
2117        eprintln!("[{label}] stdout:\n{stdout}");
2118    }
2119    if !stderr.trim().is_empty() {
2120        eprintln!("[{label}] stderr:\n{stderr}");
2121    }
2122    process::exit(output.status.code().unwrap_or(1));
2123}
2124
2125fn collect_structural_eval_runs(dir: &Path) -> Vec<harn_vm::orchestration::RunRecord> {
2126    let mut paths: Vec<PathBuf> = fs::read_dir(dir)
2127        .unwrap_or_else(|error| {
2128            command_error(&format!(
2129                "failed to read structural eval run dir {}: {error}",
2130                dir.display()
2131            ))
2132        })
2133        .filter_map(|entry| entry.ok().map(|entry| entry.path()))
2134        .filter(|entry| entry.extension().and_then(|ext| ext.to_str()) == Some("json"))
2135        .collect();
2136    paths.sort();
2137    let mut runs: Vec<_> = paths
2138        .iter()
2139        .map(|path| load_run_record_or_exit(path))
2140        .collect();
2141    runs.sort_by(|left, right| {
2142        (
2143            left.started_at.as_str(),
2144            left.workflow_id.as_str(),
2145            left.task.as_str(),
2146        )
2147            .cmp(&(
2148                right.started_at.as_str(),
2149                right.workflow_id.as_str(),
2150                right.task.as_str(),
2151            ))
2152    });
2153    runs
2154}
2155
2156/// Parse a .harn file, returning (source, AST). Exits on error.
2157pub(crate) fn parse_source_file(path: &str) -> (String, Vec<harn_parser::SNode>) {
2158    let source = match fs::read_to_string(path) {
2159        Ok(s) => s,
2160        Err(e) => {
2161            eprintln!("Error reading {path}: {e}");
2162            process::exit(1);
2163        }
2164    };
2165
2166    let mut lexer = Lexer::new(&source);
2167    let tokens = match lexer.tokenize() {
2168        Ok(t) => t,
2169        Err(e) => {
2170            let diagnostic = harn_parser::diagnostic::render_diagnostic(
2171                &source,
2172                path,
2173                &error_span_from_lex(&e),
2174                "error",
2175                &e.to_string(),
2176                Some("here"),
2177                None,
2178            );
2179            eprint!("{diagnostic}");
2180            process::exit(1);
2181        }
2182    };
2183
2184    let mut parser = Parser::new(tokens);
2185    let program = match parser.parse() {
2186        Ok(p) => p,
2187        Err(err) => {
2188            if parser.all_errors().is_empty() {
2189                let span = error_span_from_parse(&err);
2190                let diagnostic = harn_parser::diagnostic::render_diagnostic(
2191                    &source,
2192                    path,
2193                    &span,
2194                    "error",
2195                    &harn_parser::diagnostic::parser_error_message(&err),
2196                    Some(harn_parser::diagnostic::parser_error_label(&err)),
2197                    harn_parser::diagnostic::parser_error_help(&err),
2198                );
2199                eprint!("{diagnostic}");
2200            } else {
2201                for e in parser.all_errors() {
2202                    let span = error_span_from_parse(e);
2203                    let diagnostic = harn_parser::diagnostic::render_diagnostic(
2204                        &source,
2205                        path,
2206                        &span,
2207                        "error",
2208                        &harn_parser::diagnostic::parser_error_message(e),
2209                        Some(harn_parser::diagnostic::parser_error_label(e)),
2210                        harn_parser::diagnostic::parser_error_help(e),
2211                    );
2212                    eprint!("{diagnostic}");
2213                }
2214            }
2215            process::exit(1);
2216        }
2217    };
2218
2219    (source, program)
2220}
2221
2222fn error_span_from_lex(e: &harn_lexer::LexerError) -> harn_lexer::Span {
2223    match e {
2224        harn_lexer::LexerError::UnexpectedCharacter(_, span)
2225        | harn_lexer::LexerError::UnterminatedString(span)
2226        | harn_lexer::LexerError::UnterminatedBlockComment(span) => *span,
2227    }
2228}
2229
2230fn error_span_from_parse(e: &harn_parser::ParserError) -> harn_lexer::Span {
2231    match e {
2232        harn_parser::ParserError::Unexpected { span, .. } => *span,
2233        harn_parser::ParserError::UnexpectedEof { span, .. } => *span,
2234    }
2235}
2236
2237/// Execute source code and return the output. Used by REPL and conformance tests.
2238pub(crate) async fn execute(source: &str, source_path: Option<&Path>) -> Result<String, String> {
2239    let mut lexer = Lexer::new(source);
2240    let tokens = lexer.tokenize().map_err(|e| e.to_string())?;
2241    let mut parser = Parser::new(tokens);
2242    let program = parser.parse().map_err(|e| e.to_string())?;
2243
2244    // Static cross-module resolution: when executed from a file, derive the
2245    // import graph so `execute` catches undefined calls at typecheck time.
2246    // The REPL / `-e` path invokes this without `source_path`, where there
2247    // is no importing file context; we fall back to no-imports checking.
2248    let mut checker = TypeChecker::new();
2249    if let Some(path) = source_path {
2250        let graph = harn_modules::build(&[path.to_path_buf()]);
2251        if let Some(imported) = graph.imported_names_for_file(path) {
2252            checker = checker.with_imported_names(imported);
2253        }
2254        if let Some(imported) = graph.imported_type_declarations_for_file(path) {
2255            checker = checker.with_imported_type_decls(imported);
2256        }
2257    }
2258    let type_diagnostics = checker.check(&program);
2259    let mut warning_lines = Vec::new();
2260    for diag in &type_diagnostics {
2261        match diag.severity {
2262            DiagnosticSeverity::Error => return Err(diag.message.clone()),
2263            DiagnosticSeverity::Warning => {
2264                warning_lines.push(format!("warning: {}", diag.message));
2265            }
2266        }
2267    }
2268
2269    let chunk = harn_vm::Compiler::new()
2270        .compile(&program)
2271        .map_err(|e| e.to_string())?;
2272
2273    let local = tokio::task::LocalSet::new();
2274    local
2275        .run_until(async {
2276            let mut vm = harn_vm::Vm::new();
2277            harn_vm::register_vm_stdlib(&mut vm);
2278            install_default_hostlib(&mut vm);
2279            let source_parent = source_path
2280                .and_then(|p| p.parent())
2281                .unwrap_or(std::path::Path::new("."));
2282            let project_root = harn_vm::stdlib::process::find_project_root(source_parent);
2283            let store_base = project_root.as_deref().unwrap_or(source_parent);
2284            let execution_cwd = std::env::current_dir()
2285                .unwrap_or_else(|_| std::path::PathBuf::from("."))
2286                .to_string_lossy()
2287                .into_owned();
2288            let source_dir = source_parent.to_string_lossy().into_owned();
2289            if source_path.is_some_and(is_conformance_path) {
2290                harn_vm::event_log::install_memory_for_current_thread(64);
2291            }
2292            harn_vm::register_store_builtins(&mut vm, store_base);
2293            harn_vm::register_metadata_builtins(&mut vm, store_base);
2294            let pipeline_name = source_path
2295                .and_then(|p| p.file_stem())
2296                .and_then(|s| s.to_str())
2297                .unwrap_or("default");
2298            harn_vm::register_checkpoint_builtins(&mut vm, store_base, pipeline_name);
2299            harn_vm::stdlib::process::set_thread_execution_context(Some(
2300                harn_vm::orchestration::RunExecutionRecord {
2301                    cwd: Some(execution_cwd),
2302                    source_dir: Some(source_dir),
2303                    env: std::collections::BTreeMap::new(),
2304                    adapter: None,
2305                    repo_path: None,
2306                    worktree_path: None,
2307                    branch: None,
2308                    base_ref: None,
2309                    cleanup: None,
2310                },
2311            ));
2312            if let Some(ref root) = project_root {
2313                vm.set_project_root(root);
2314            }
2315            if let Some(path) = source_path {
2316                if let Some(parent) = path.parent() {
2317                    if !parent.as_os_str().is_empty() {
2318                        vm.set_source_dir(parent);
2319                    }
2320                }
2321            }
2322            // Conformance tests land here via `run_conformance_tests`; for
2323            // `skill_fs_*` fixtures to see the bundled `skills/` folder
2324            // we run the same layered discovery as `harn run`.
2325            let loaded = skill_loader::load_skills(&skill_loader::SkillLoaderInputs {
2326                cli_dirs: Vec::new(),
2327                source_path: source_path.map(Path::to_path_buf),
2328            });
2329            skill_loader::emit_loader_warnings(&loaded.loader_warnings);
2330            skill_loader::install_skills_global(&mut vm, &loaded);
2331            if let Some(path) = source_path {
2332                let extensions = package::load_runtime_extensions(path);
2333                package::install_runtime_extensions(&extensions);
2334                package::install_manifest_triggers(&mut vm, &extensions)
2335                    .await
2336                    .map_err(|error| format!("failed to install manifest triggers: {error}"))?;
2337                package::install_manifest_hooks(&mut vm, &extensions)
2338                    .await
2339                    .map_err(|error| format!("failed to install manifest hooks: {error}"))?;
2340            }
2341            let _event_log = harn_vm::event_log::active_event_log()
2342                .unwrap_or_else(|| harn_vm::event_log::install_memory_for_current_thread(64));
2343            let connector_clients_installed =
2344                should_install_default_connector_clients(source, source_path);
2345            if connector_clients_installed {
2346                install_default_connector_clients(store_base)
2347                    .await
2348                    .map_err(|error| format!("failed to initialize connector clients: {error}"))?;
2349            }
2350            let execution_result = vm.execute(&chunk).await.map_err(|e| e.to_string());
2351            harn_vm::egress::reset_egress_policy_for_host();
2352            if connector_clients_installed {
2353                harn_vm::clear_active_connector_clients();
2354            }
2355            harn_vm::stdlib::process::set_thread_execution_context(None);
2356            execution_result?;
2357            let mut output = String::new();
2358            for wl in &warning_lines {
2359                output.push_str(wl);
2360                output.push('\n');
2361            }
2362            output.push_str(vm.output());
2363            Ok(output)
2364        })
2365        .await
2366}
2367
2368fn should_install_default_connector_clients(source: &str, source_path: Option<&Path>) -> bool {
2369    if !source_path.is_some_and(is_conformance_path) {
2370        return true;
2371    }
2372    source.contains("connector_call")
2373        || source.contains("std/connectors")
2374        || source.contains("connectors/")
2375}
2376
2377fn is_conformance_path(path: &Path) -> bool {
2378    path.components()
2379        .any(|component| component.as_os_str() == "conformance")
2380}
2381
2382async fn install_default_connector_clients(base_dir: &Path) -> Result<(), String> {
2383    let event_log = harn_vm::event_log::active_event_log()
2384        .unwrap_or_else(|| harn_vm::event_log::install_memory_for_current_thread(64));
2385    let secret_namespace = connector_secret_namespace(base_dir);
2386    let secrets: Arc<dyn harn_vm::secrets::SecretProvider> = Arc::new(
2387        harn_vm::secrets::configured_default_chain(secret_namespace)
2388            .map_err(|error| format!("failed to configure secret providers: {error}"))?,
2389    );
2390
2391    let registry = harn_vm::ConnectorRegistry::default();
2392    let metrics = Arc::new(harn_vm::MetricsRegistry::default());
2393    let inbox = Arc::new(
2394        harn_vm::InboxIndex::new(event_log.clone(), metrics.clone())
2395            .await
2396            .map_err(|error| error.to_string())?,
2397    );
2398    registry
2399        .init_all(harn_vm::ConnectorCtx {
2400            event_log,
2401            secrets,
2402            inbox,
2403            metrics,
2404            rate_limiter: Arc::new(harn_vm::RateLimiterFactory::default()),
2405        })
2406        .await
2407        .map_err(|error| error.to_string())?;
2408    let clients = registry.client_map().await;
2409    harn_vm::install_active_connector_clients(clients);
2410    Ok(())
2411}
2412
2413fn connector_secret_namespace(base_dir: &Path) -> String {
2414    match std::env::var("HARN_SECRET_NAMESPACE") {
2415        Ok(namespace) if !namespace.trim().is_empty() => namespace,
2416        _ => {
2417            let leaf = base_dir
2418                .file_name()
2419                .and_then(|name| name.to_str())
2420                .filter(|name| !name.is_empty())
2421                .unwrap_or("workspace");
2422            format!("harn/{leaf}")
2423        }
2424    }
2425}
2426
2427#[cfg(test)]
2428mod main_tests {
2429    use super::{normalize_serve_args, should_install_default_connector_clients};
2430    use std::path::Path;
2431
2432    #[test]
2433    fn normalize_serve_args_inserts_a2a_for_legacy_shape() {
2434        let args = normalize_serve_args(vec![
2435            "harn".to_string(),
2436            "serve".to_string(),
2437            "--port".to_string(),
2438            "3000".to_string(),
2439            "agent.harn".to_string(),
2440        ]);
2441        assert_eq!(
2442            args,
2443            vec![
2444                "harn".to_string(),
2445                "serve".to_string(),
2446                "a2a".to_string(),
2447                "--port".to_string(),
2448                "3000".to_string(),
2449                "agent.harn".to_string(),
2450            ]
2451        );
2452    }
2453
2454    #[test]
2455    fn normalize_serve_args_preserves_explicit_subcommands() {
2456        let args = normalize_serve_args(vec![
2457            "harn".to_string(),
2458            "serve".to_string(),
2459            "acp".to_string(),
2460            "server.harn".to_string(),
2461        ]);
2462        assert_eq!(
2463            args,
2464            vec![
2465                "harn".to_string(),
2466                "serve".to_string(),
2467                "acp".to_string(),
2468                "server.harn".to_string(),
2469            ]
2470        );
2471    }
2472
2473    #[test]
2474    fn conformance_skips_connector_clients_unless_fixture_uses_connectors() {
2475        let path = Path::new("conformance/tests/language/basic.harn");
2476        assert!(!should_install_default_connector_clients(
2477            "println(1)",
2478            Some(path)
2479        ));
2480        assert!(!should_install_default_connector_clients(
2481            "trust_graph_verify_chain()",
2482            Some(path)
2483        ));
2484        assert!(should_install_default_connector_clients(
2485            "import { post_message } from \"std/connectors/slack\"",
2486            Some(path)
2487        ));
2488        assert!(should_install_default_connector_clients(
2489            "println(1)",
2490            Some(Path::new("examples/demo.harn"))
2491        ));
2492    }
2493}
harn_cli/lib.rs

harn_cli/
lib.rs
