trigger a verification run with aviator verify r/N

[simsinght]

Runbook: https://app.aviator.co/r/1215

aviator verify gains a trigger mode: aviator verify r/N starts a verification run on an existing verify session — including the session's first run (deliberately not a rerun subcommand). Bare aviator verify keeps its submit behavior unchanged.

• --force — fresh full run even when the server would dedup against an equivalent existing run (dedup bypass only, like the webapp's rerun button; a run already in progress still refuses).
• --evaluator-only — re-judge already-collected evidence instead of re-driving the collector (the cheap path after a criteria edit); bypasses dedup by design.
• --json — stable-key object (runbook_number, runbook_id, url, run_id, run_status, deduplicated, message).
• Server dedup (200 + deduplicated: true) is reported as success, so agents can call this liberally after every push.
• Each mode rejects the other mode's flags by name; --repo/--intent required-ness moved from cobra into submit-mode checks.

Client half of the mergeit change that decouples criteria edits from verification runs and adds POST /api/v1/verify/<n>/run (branch hold-verify-run-on-ac-edit); the two should land together, backend first.

https://claude.ai/code/session_015gofgfqntp3kWNWimXYvv6

[aviator-app]

Current Aviator status

Aviator will automatically update this comment as the status of the PR changes.
Comment /aviator refresh to force Aviator to re-examine your PR (or learn about other /aviator commands).

This PR was merged using Aviator (commit e9a854f).

Stack

• #22 add --regen-scenarios to aviator verify r/N
• 👉 #21 trigger a verification run with aviator verify r/N (this PR)
• master

---

See the real-time status of this PR on the Aviator webapp.

Use the Aviator Chrome Extension to see the status of your PR within GitHub.

[aviator-app]

Connected this PR to r/1215.

[aviator-app]

✅ FlexReview Status

Common Owner: aviator-co/engineering (expert-load-balance assignment)
Owner and Assignment:

• ✅ aviator-co/engineering (expert-load-balance assignment)
  • Slack: #fr-eng-notifications
  • Assigned: brain-crystal, jainankit
  Owned Files

  • 🔒 cmd/aviator/verify.go
  • 🔒 cmd/aviator/verify_test.go
  • 🔒 internal/api/verify_run.go
  • 🔒 internal/api/verify_run_test.go

Review SLO: 7 business hours if PR size is <= 200 LOC for the first response.
❕ This PR modifies 421 lines, which is larger than the Review SLO threshold.

[aviator-app]

✅ Verification passed · 0fcd243

5/5 passed

Per-criterion results

✅ 1. aviator verify r/ (bare number and runbook URL also accepted) triggers a verification run on that existing session — which may be the session's first run — and prints the run status, the session URL, and the server's message, exiting 0.

✅ 2. When the server deduplicates (an equivalent non-error run already exists for the current commit and criteria set), the command exits 0 and states that an existing equivalent run was returned rather than claiming a new run started.

✅ 3. --force and --evaluator-only are carried to the server in the trigger request (force / evaluator_only), so a forced fresh run or an evaluator-only re-grade can be requested from the CLI.

✅ 4. --json in trigger mode emits exactly one JSON object on stdout with the stable keys runbook_number, runbook_id, url, run_id, run_status, deduplicated, and message, with no human-formatted text mixed into stdout.

✅ 5. Mode mixing is rejected with an error naming the offending flag: submission-only flags (--repo, --intent, --criteria, and friends) with an r/ argument, and --force or --evaluator-only without one; bare aviator verify submission behavior is otherwise unchanged.

View in Aviator