gh-ghent（根特）

gh-ghent — Agentic PR Monitoring

Prerequisite: gh extension install indrasvat/gh-ghent

All commands require: --pr <N> --format json --no-tui

Get PR number: gh pr view --json number -q.number

First Command After PR Creation Or Review-Fix Push

PR=$(gh pr view --json number -q .number)
gh ghent status --pr $PR --await-review --solo --logs --format json --no-tui

This is the single blessed command for PR review handling. Use it:

• immediately after PR creation
• again after every push that addresses review or CI feedback

It waits for CI, performs bounded review monitoring, and returns everything in one response:

• threads with is_bot
• checks with log excerpts
• reviews
• review_monitor
• is_merge_ready

Drop --solo for org repos with required review policies. Always include --await-review when review comments may still arrive. Do not switch to bare --watch after the first cycle if review comments still matter — --watch is CI-only and can miss follow-up bot comments. Drop --logs only on narrow re-checks where CI failure detail is definitely not needed.

Response Shape (status)

{
  "is_merge_ready": false,
  "comments": {
    "threads": [{"id": "PRRT_...", "path": "foo.go", "line": 42,
                  "comments": [{"author": "coderabbitai", "is_bot": true, "body": "..."}]}],
    "unresolved_count": 2,
    "bot_thread_count": 2,
    "unanswered_count": 1
  },
  "checks": {
    "overall_status": "pass",
    "checks": [{"name": "CI", "conclusion": "success", "annotations": [], "log_excerpt": ""}]
  },
  "reviews": [{"author": "alice", "state": "APPROVED"}],
  "review_monitor": {
    "phase": "settled",
    "confidence": "high",
    "activity_count": 3,
    "wait_seconds": 154
  }
}

Decision Order

Act on the first matching condition — fix it, then re-run status:

1. Exit code 2 → auth / rate limit / not-found error. Fix credentials.
2. checks.overall_status == "failure" → Fix CI. Log excerpts and annotations are inline.
3. checks.overall_status == "pending" → Re-run the same status --await-review command. Do not switch to --watch while review comments may still appear.
4. comments.unanswered_count > 0 → Bot sweep (see below).
5. comments.unresolved_count > 0 → gh ghent resolve --pr <N> --all
6. review_monitor.phase == "timeout" or review_monitor.confidence == "low" → Treat result as provisional. If you just pushed fixes, re-run the same status --await-review command after the push settles.
7. is_merge_ready == true and review_monitor.confidence!= "low" → Merge / stop.

Anti-Footgun Rule

When review comments may still arrive:

• use gh ghent status --await-review...
• after every fix push, use gh ghent status --await-review... again
• do not switch to gh ghent checks --watch
• do not switch to gh ghent status --watch

Bare --watch is only for CI-only waiting when review state is irrelevant.

Bot Sweep (when unanswered_count > 0)

The status already contains the full threads — no second call needed.

1. Read threads from comments.threads[] where comments[0].is_bot == true
2. Fix code → push
3. Per thread: gh ghent reply --pr <N> --thread PRRT_... --body "Fixed" --resolve
4. Re-check with the same command: gh ghent status --pr <N> --await-review --solo --logs --format json --no-tui
5. Repeat until is_merge_ready == true and review_monitor.confidence!= "low"

Solo Mode

Add --solo only when the repo owner is the authenticated user on a personal (non-org) repo. Never auto-add for org repos. If merge readiness is false only because approval is missing on a personal repo, retry with --solo.

Commands

Exit Codes

Exit 2 = auth failure, rate limit, or resource not found.

Other Patterns

# Merge-readiness gate (silent exit 0 if ready, exit 1 + full output if not)
gh ghent status --pr <N> --quiet --solo

# Drill-down: bot threads only
gh ghent comments --pr <N> --bots-only --unanswered --format json --no-tui

# Group by file for batch fixing
gh ghent comments --pr <N> --group-by file --format json --no-tui

# Compact status (minimal tokens for polling loops)
gh ghent status --pr <N> --compact --format json --no-tui

References

• Command Reference — all flags, full output schemas
• Agent Workflows — step-by-step patterns
• Exit Codes — branching logic
• Review Cycle Example — read, fix, resolve, reply walkthrough
• CI Monitor Example — watch CI, extract errors, fix, re-check