jackin-sentinel

Internal jackin' role for manual smoke checks and future PTY, functional, and Docker E2E coverage.

The canonical manual-test repository is:

https://github.com/jackin-project/jackin-sentinel

Load it from any workspace with:

jackin load jackin-project/sentinel . --rebuild --debug

The role intentionally concentrates the valid role-manifest surface area in one place:

• every supported agent table
• runtime hooks: setup_once, source, and preflight
• static env defaults
• rich text prompts
• rich select prompts
• skippable prompts and skip cascading
• dependency ordering and ${env.*} interpolation
• Docker build output from a real RUN layer

At runtime the hooks install jackin-sentinel-report into the agent user's PATH. Run it from the container to print a deterministic report bounded by JACKIN_SENTINEL_REPORT_BEGIN and JACKIN_SENTINEL_REPORT_END. Future E2E tests should parse that report rather than depending on agent-specific UI text.

Manual verification checklist:

1. Confirm the launch cockpit owns the full screen and no raw Docker output is printed over it.
2. Complete every manifest env dialog: required text, defaulted text, select, defaulted select, skippable optional value, and dependency-derived values.
3. Open the Docker build log from the cockpit and confirm the jackin-sentinel build layer line appears there, not on the main screen.
4. Launch at least one agent and run jackin-sentinel-report.
5. Confirm the report shows resolved env values, hook markers, runtime variables, and Docker CLI availability.
6. Exit the container with a dirty isolated worktree and confirm cleanup is a rich dialog, not a plain numbered CLI prompt.

Recommended test tiers:

1. PTY tests drive the rich launch dialogs and assert the visible screen.
2. Functional tests run jackin' commands against this role with fake or stubbed external dependencies.
3. Docker E2E tests launch the role with a real Docker daemon and assert the sentinel report, hook markers, env resolution, and diagnostics/build output.