From gosmith
Fetch PR review comments, triage against codebase, apply fixes, generate reply report.
npx claudepluginhub otaviof/gosmith --plugin gosmithThis skill is limited to using the following tools:
Address PR review comments using MCRF with progressive confidence tracking.
Verifies tests pass on completed feature branch, presents options to merge locally, create GitHub PR, keep as-is or discard; executes choice and cleans up worktree.
Guides root cause investigation for bugs, test failures, unexpected behavior, performance issues, and build failures before proposing fixes.
Writes implementation plans from specs for multi-step tasks, mapping files and breaking into TDD bite-sized steps before coding.
Address PR review comments using MCRF with progressive confidence tracking.
DECOMPOSE → Fetch comments, detect sources (Phase 1)
SOLVE → Categorize by action required (Phase 2)
VERIFY → Read code, validate claims (Phase 3)
SYNTHESIZE → Generate reply report (Phase 4)
REFLECT → Final confidence + caveats
Detect repo owner/name and fetch all comment sources:
# Repo metadata
gh repo view --json nameWithOwner --jq '.nameWithOwner'
# PR overview
gh pr view $pr_number --json title,body,state,comments,reviews,reviewDecision
# Inline review comments
gh api repos/{owner}/{repo}/pulls/$pr_number/comments \
--jq '.[] | {id, path, line, body, user: .user.login}'
# Review-level comments
gh api repos/{owner}/{repo}/pulls/$pr_number/reviews \
--jq '.[] | {id, user: .user.login, state, body}'
Report: Count by source, parse confidence per comment.
CodeRabbit embeds structured feedback inside review bodies using nested <details> HTML blocks:
_Potential issue_ and severity badges<summary>Nitpick comments</summary> blocks, contain file path and line ranges<summary>Outside diff range comments</summary> blocks<summary>Prompt for AI Agents</summary> — use these as implementation hints but always verify against code firstHuman patterns: Short, conversational. "food for thought", "not a change on this PR", questions → typically observations not action requests.
Categorize each comment:
| Category | Criteria | Action |
|---|---|---|
| Actionable | Bugs, factual errors, misleading docs, test gaps | Fix |
| Food for thought | Observations, future ideas, questions | Acknowledge |
| Out of scope | Requires changes beyond PR purpose | Acknowledge + explain |
Present triage table for user approval before proceeding:
| # | Source | File | What | Why | Category | Confidence |
|---|
Wait for user confirmation.
Track with TodoWrite. For each actionable item:
Constraints: No commits, no GitHub API posts, only working tree edits.
One section per comment with GitHub links:
https://github.com/{owner}/{repo}/pull/{pr}#discussion_r{id}https://github.com/{owner}/{repo}/pull/{pr}#pullrequestreview-{id}Format:
### [Comment 1](link) — User: description ([file:L42](path#L42))
Confidence: 0.95
` ` `markdown
Fixed — what was done.
` ` `
---
### [Comment 2](link) — User: review nitpicks
Confidence: 0.90
` ` `markdown
All N items addressed:
- Item 1 description
- Item 2 description
` ` `
Grouping: Combine review-body nitpicks (CodeRabbit or human) under single section using review ID.