Re-request reviews after addressing changes

[dgershman]

Summary

• Adds a re-request review instruction to the changesRequested auto-respond prompt in AutoRespondCoordinator
• After Claude Code addresses review comments and pushes, it now also runs gh pr edit --add-reviewer <login> (GitHub) or glab mr update --reviewer <login> (GitLab) to re-request review from each reviewer who requested changes
• Both the auto-respond and manual "Address Changes" quick action paths pick up the change since they share AutoRespondPrompts.build()

Closes #275

Test plan

• Trigger a "changes requested" review on a PR managed by Crow
• Verify the injected prompt includes the re-request instruction
• Verify Claude Code fetches reviews, addresses comments, pushes, then runs gh pr edit --add-reviewer
• Verify the reviewer receives a GitHub notification that their review was re-requested

🤖 Generated with Claude Code

[dhilgaertner]

Code & Security Review

Critical Issues

None — change is a 4-line string augmentation to the changesRequested auto-respond prompt.

Security Review

Strengths:

• No new shell invocation surface in Crow itself; the appended hint is text injected into a managed terminal for Claude Code to act on. transition.prURL is already interpolated into the existing fetchHint in the same string, so this introduces no new injection vector.
• No new API scopes, secrets, or auth flows. Consistent with the file's stated design (Crow does not fetch review bodies or CI logs itself).
• Manual addressChanges quick-action path inherits the change for free via AutoRespondPrompts.build() (Sources/Crow/App/AutoRespondCoordinator.swift:168), avoiding drift between auto and manual flows.

Concerns:

• None.

Code Quality

• Yellow — glab mr update --reviewer semantics (Sources/Crow/App/AutoRespondCoordinator.swift:122): Unlike gh pr edit --add-reviewer, the GitLab glab mr update --reviewer <login> flag sets the reviewer list rather than appending. Running it once per reviewer (as the prompt instructs: "for each one") would only leave the last login as the reviewer, dropping the others. Consider either (a) instructing Claude to pass a comma-separated list in a single invocation, or (b) noting the set-vs-add distinction so Claude doesn't iterate. The GitHub path is fine — --add-reviewer is additive.
• Green — Prompt length (Sources/Crow/App/AutoRespondCoordinator.swift:128): The changesRequested prompt is now noticeably longer, still on a single line. Given the existing single-line + \n contract documented at the top of AutoRespondPrompts (line 106-110), this is fine, but worth keeping an eye on if more hints accumulate.
• Green — No tests: There are no tests covering AutoRespondPrompts.build() in the repo, so the substring change is verified manually only. A small snapshot-style test asserting the prompt contains both gh pr edit and --add-reviewer for GitHub and glab mr update/--reviewer for GitLab would lock down the contract cheaply. Not a blocker.

Summary Table

Priority	Issue
🟡 Yellow	glab mr update --reviewer replaces reviewers; iterating per-login will drop earlier ones (line 122)
🟢 Green	Consider a snapshot test for the prompt strings to prevent silent regressions
🟢 Green	Prompt is approaching a chunky single line — monitor as more hints are appended

Recommendation: Approve. The change is low-risk, small, and the manual quick-action path is correctly covered via shared AutoRespondPrompts.build(). The GitLab semantics nit is worth a follow-up but doesn't block merge — Claude Code can self-correct when the command behaves unexpectedly.

---

🤖 Reviewed by Crow via Claude Code