Executable independent-runner example: CI-identity review dispatch

[tylerwillis]

Why

specs/REQ-008-honest-boundaries.md (REQ-008.1.3) names the independent-runner pattern — a CI identity the author cannot impersonate re-dispatches the judgment reviews and records the verdicts — as the trust boundary, "presented as a reference architecture until an executable example ships." This issue is that example.

It's also the biggest adoption gap for fully headless pipelines: 2119 review --dispatch emits a prompt aimed at an interactive agent session, and a team wiring a scanner→PR or issue→PR pipeline with no human at a keyboard currently has to invent the dispatch step themselves. An adopter evaluating 2119 for governed enterprise workflows will make this their first question.

Shape

A working GitHub Actions workflow (in this repo or a small example repo) where:

1. A CI-owned identity — not the PR author — checks out the branch and runs 2119 review.
2. Each instruction file is executed by a fresh-context agent under that CI identity, headless.
3. Verdicts are recorded with 2119 pass/2119 fail and committed (or pushed as a commit the author cannot forge).
4. 2119 check then gates the merge as usual.

Runtime suggestion

The ai-outfitter control system already covers the hard parts: actions runs agent profiles headless in GitHub Actions on any trigger, and agent-operator runs them in-cluster. A reviewer profile running under ai-outfitter/actions with its own identity is probably the shortest path to the executable example — and demonstrates the two projects composing.

Done means

• A runnable workflow an adopter can copy.
• docs/scaling.md's independent-runner recipe links it.
• REQ-008.1.3's "until an executable example ships" clause can be retired.

🤖 Generated with Claude Code