feat: run halt command process

[a5c-ai]

Links #880

Summary

• Adds .a5c/processes/issue-880-run-halt-command.mjs as an executable Babysitter process definition for implementing babysitter run:halt.
• Adds .a5c/processes/issue-880-run-halt-command.inputs.json with concrete issue inputs, target files, tests, docs, quality commands, acceptance criteria, and non-goals.

Planned phases

• Read issue Add 'babysitter run:halt' command for sealing divergent runs #880 and related recovery context.
• Run Phase 0 reuse audit and trace CLI/journal/state call paths.
• Design the manual-seal CLI contract, including final-status semantics and audit fields.
• Author failing CLI regression tests before implementation.
• Implement the command wiring, handler, journal append, state rebuild, and docs in a bounded loop.
• Run verification, code review, documentation, and final acceptance gates.

Expected tasks

• Extend CLI registration, positionals, flag parsing/types, usage, command exports, and dispatch.
• Add a handler near existing recovery commands that uses appendEvent and rebuildStateCache.
• Record manual_seal_reason, command source, requested status, prior state, and pending-effect summary.
• Cover dry-run, JSON output, supported final statuses, already-terminal rejection, state rebuild, and visibility through status/events.

Quality gates

• npm run build:sdk
• targeted CLI run tests for packages/sdk/src/cli/__tests__/cliRuns.test.ts
• npm run test:sdk
• npm run verify:metadata
• reviewer gates for journal invariants, audit trail, completionProof semantics, docs, and scoped diff.

[a5c-ai]

Implemented the planned babysitter run:halt changes from issue #880 and pushed commit c860eefb4 to plan/issue-880.

Summary:

• Added CLI wiring, parsing, dispatch, usage, and handler support for babysitter run:halt <runDir> --reason <text> [--final-status halted|completed|failed] [--json] [--dry-run].
• Implemented manual terminal sealing via appendEvent, defaulting to RUN_HALTED, with explicit RUN_COMPLETED/RUN_FAILED support.
• Recorded audit fields including manual_seal_reason, sealedBy, requested status, prior lifecycle state/event, and pending effects summary.
• Rebuilt state with manual_seal after mutation, rejected already-terminal runs without mutation, and preserved completionProof only for completed manual seals.
• Added CLI regression coverage and updated CLI/reference/recovery docs and prompt templates.

Verification:

• npm run build:sdk passed.
• npm run test --workspace=@a5c-ai/babysitter-sdk -- --run src/cli/__tests__/cliRuns.test.ts passed: 62 tests.
• npm run test:sdk passed: 2080 tests, 9 skipped.
• npm run verify:metadata still fails due to unrelated pre-existing .agents/plugins/marketplace.json metadata: expected babysitter version 5.0.1-staging.28347f861706, found undefined. That file was left unstaged.

[a5c-ai]

Implemented the planned run halt process on plan/issue-880.

Summary:

• Added babysitter run:halt <runDir> --reason <text> [--final-status halted|completed|failed] [--json] [--dry-run].
• Wires command parsing, dispatch, usage, and docs for manual run sealing.
• Appends audited terminal events via appendEvent, records manual seal reason/source/prior state/pending effects, and rebuilds state with manual_seal.
• Added CLI regression coverage for required reason, invalid statuses, dry-run behavior, default halted seal, failed/completed seals, terminal-run rejection, state rebuild, and event visibility.

Verification:

• git diff --check passed for the scoped files.
• npm run build:sdk passed.
• npm run test:sdk -- --run src/cli/__tests__/cliRuns.test.ts passed.
• npm run test:sdk passed: 146 files passed, 1 skipped; 2080 tests passed, 9 skipped.
• npm run verify:metadata failed on pre-existing local metadata drift in .agents/plugins/marketplace.json: expected babysitter version 5.0.1-staging.28347f861706, found undefined. That file is unrelated to this implementation and was left unstaged.

Commit pushed: c860eefb4 feat(sdk): add run halt command.