shape-review-comments
Shapes raw review findings into postable file:line comments written in Mihai's voice.
Skill by MihaiOnSoftware
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/mihaionsoftware-shape-review-comments
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/shape-review-comments
What's inside
- Skill name
shape-review-comments- Description
- Shapes raw review findings into postable file:line comments written in Mihai's voice. Use when you have findings from an adversarial-review-loop or mihai-code-review run and need to decide what to post, what to drop, and how to phrase what survives.
- License
- MIT
- Instructions
- 55 lines in SKILL.md · view on GitHub
Files (3)
- SKILL.md
- triage-rules.md · references/
- voice-guide.md · references/
SKILL.md (first lines)
Takes raw review findings and produces postable comments: triage what to skip, frame what survives in Mihai's voice at the right file:line, verify line numbers, then present for approval. Reference files (load both before writing a single comment): - [references/triage-rules.md](references/triage-rules.md) — what to drop, what to keep, severity labels - [references/voice-guide.md](references/voice-guide.md) — sentence structure, hedges, closers, worked examples If either reference file cannot be read, stop and report it — do not attempt to triage or write comments from memory. If the findings list is empty, tell the user and stop. ## Step 1 — Load the references Read both reference files completely before proceeding. ## Step 2 — Triage For every finding, assign one of: - **Post** — survives, worth saying - **Drop** — not worth saying; note the reason briefly so it isn't reconsidered - **Merge** — combine with a closely related finding on the same line or function Apply triage rules before writing any comment text. Do not let wording decisions influence whether a finding survives. Produce a visible triage log (one line per finding: decision + brief reason) before moving to Step 3. This lets the user catch wrong drops before comments are written. ## Step 3 — Write the comments For each surviving finding, write the comment following the voice guide (anchor line selection, hedges, closers, severity labeling are covered there). Two hard rules apply to every comment before you move on: 1. **No em-dashes.** Do not write them. A comma or a new sentence always works. 2. **Maximum 2 sentences.** If a third sentence feels necessary, cut it. If that loses the finding, rewrite from the observation, not the explanation. Output format: ``` `path/to/file.ts:LINE` — severity: comment text ``` ## Step 4 — Verify line numbers Before presenting to the user, confirm each line number at current HEAD. Run `grep -n` for the anchoring expression in the file. If the file has changed since the review and a finding was already fixed, drop that comment and note it. While here, verify any cross-file references inside comment bodies too, and convert them to GitHub links at the PR's head commit (see the voice guide's Precedent links section for the URL format). If a line number cannot be confirmed (file deleted, path changed), drop the comment and tell the user. ## Step 5 — Present for approval Present the shaped comments to the user. Walk through them one at a time if there are many, or offer to batch them if they are straightforward. If the user rejects all comments, ask whether to re-triage from scratch with different criteria or to close. Do not loop silently.
From MihaiOnSoftware/.ai, MIT.
More from MihaiOnSoftware/.ai
- 0.592score
review-pipeline
MihaiOnSoftware
Parallel code review pipeline: run adversarial-review-loop and mihai-code-review simultaneously, merge their findings, shape into postable comments, and…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
mihai-code-review
MihaiOnSoftware
Review a PR, branch, or diff the way Mihai would: mechanical style/test sweeps first, then precedent diffing, then an adversarial correctness hunt.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
monitor-gh-pr
MihaiOnSoftware
Monitor a GitHub PR for CI failures, CI success, and new comments (including inline review comments).
SkillAFreshness: A★ 1Claude Code +9 - 0.502score
create-implementation-plan
MihaiOnSoftware
Break a large task into incremental slices using story splitting patterns.
SkillCFreshness: C★ 1Claude Code +9 - 0.502score
adversarial-review-loop
MihaiOnSoftware
Run fresh adversarial subagent, incorporate findings, repeat until findings are weak/repeated/invalid.
SkillCFreshness: C★ 1Claude Code +9 - 0.499score
adversarial-review
MihaiOnSoftware
Spawn a fresh-context subagent that is told a mistake exists in your work and tasked with finding it.
SkillCFreshness: C★ 1Claude Code +9 - 0.499score
solve-this-problem
MihaiOnSoftware
End-to-end: problem → design → plan → TDD slices.
SkillCFreshness: C★ 1Claude Code +9 - 0.496score
tdd-slice
MihaiOnSoftware
Implement a complete TDD slice with multiple micro-tdd cycles and validation
SkillCFreshness: C★ 1Claude Code +9 - 0.495score
explore-and-design
MihaiOnSoftware
Investigate, scope, and shape a solution before breaking it into implementation slices.
SkillCFreshness: C★ 1Claude Code +9 - 0.493score
write-validation-report
MihaiOnSoftware
Write a validation report with pass/fail verdict
SkillCFreshness: C★ 1Claude Code +9 - 0.493score
write-agent-report
MihaiOnSoftware
Write an agent report with proper naming and location
SkillCFreshness: C★ 1Claude Code +9 - 0.491score
commit
MihaiOnSoftware
Create a commit following message rules and writing style guidelines
SkillCFreshness: C★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 21, 2026 (8 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/MihaiOnSoftware/.ai
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from MihaiOnSoftware
- 0.592score
mihai-code-review
MihaiOnSoftware
Review a PR, branch, or diff the way Mihai would: mechanical style/test sweeps first, then precedent diffing, then an adversarial correctness hunt.
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
review-pipeline
MihaiOnSoftware
Parallel code review pipeline: run adversarial-review-loop and mihai-code-review simultaneously, merge their findings, shape into postable comments, and…
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
monitor-gh-pr
MihaiOnSoftware
Monitor a GitHub PR for CI failures, CI success, and new comments (including inline review comments).
SkillAFreshness: A★ 1Claude Code +9 - 0.502score
adversarial-review-loop
MihaiOnSoftware
Run fresh adversarial subagent, incorporate findings, repeat until findings are weak/repeated/invalid.
SkillCFreshness: C★ 1Claude Code +9 - 0.502score
create-implementation-plan
MihaiOnSoftware
Break a large task into incremental slices using story splitting patterns.
SkillCFreshness: C★ 1Claude Code +9 - 0.499score
adversarial-review
MihaiOnSoftware
Spawn a fresh-context subagent that is told a mistake exists in your work and tasked with finding it.
SkillCFreshness: C★ 1Claude Code +9
Cite this page
shape-review-comments skill. Agent Mart. https://agentmart.dev/skills/mihaionsoftware-shape-review-comments (last updated 2026-10-24)