grill-me-comments
Work through review comments — the user's own, a Pull Request's threads, or feedback they paste — settling the obvious ones on a single OK, grilling the rest…
Skill by javierponferradalopez
Chalk Labs
Need help deploying grill-me-comments?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
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/javierponferradalopez-grill-me-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
Use cases: backend
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
- Runs as
- Shell scripts
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/grill-me-comments
What's inside
- Skill name
grill-me-comments- Description
- Work through review comments — the user's own, a Pull Request's threads, or feedback they paste — settling the obvious ones on a single OK, grilling the rest one at a time, and handing back the agreed batch for the user to say what becomes of it.
- License
- MIT
- Instructions
- 56 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (3)
- PR.md
- SKILL.md
- pr-threads.sh · scripts/
SKILL.md (first lines)
# Grill me on the review comments Some comments are one instruction with one obvious implementation; others hide a decision that belongs to the user. Applying everything silently gets the second kind wrong, and discussing everything wastes the user on the first — so sort them, then treat each kind on its own terms. The whole run is agreement, not work: nothing is written until the user has said what becomes of the batch — not a line of code, not a reply on GitHub, not a comment deleted. A comment is closed by the change that satisfies it actually landing, so a comment nobody implemented stays exactly where it is. ## 1. Find them Start from the source the user's message already names. Invoked with nothing, ask — closed, so one word answers it: **yours, on a Pull Request, or pasted here?** - **Yours** — comments they left for you outside this conversation. Where those live and what closing one means is theirs to define: their instructions for you say it, and when nothing does, ask. Either way that is the current word, over anything here. - **On a Pull Request** — [`PR.md`](PR.md) has the ground check, the collection script and the reply mechanics. - **Pasted here** — take them as given. Read the source again before the batch of step 5 and before anything is written in step 6. The run takes turns, and the user keeps working through them: what you collected here is a copy, and a copy goes stale. ## 2. Sort — the uniqueness test For each comment ask: **is there more than one reasonable way to satisfy this?** - No → **AFK**. It resolves with the agent alone. - Yes → **HITL** — human in the loop. It only resolves through a live exchange with the user, who speaks for themselves; an agent that answers for them has broken this. A comment is HITL when it reaches code outside its own file, changes a signature or any other public contract, names a new concept, or asks a question instead of giving an instruction. So is one whose code you cannot pin down with certainty — an anchor that has drifted makes every reading of the comment a guess, whatever it asks. Borderline counts as HITL: a question costs one turn, a wrong guess costs the user's trust in the whole batch. ## 3. Post the map, then get an OK on the AFK ones Lead with the shape of the batch — _"12 comments: 7 AFK, 5 HITL."_ List the AFK ones, one line each: what the comment asks, and what you will do about it. Then wait. AFK says the implementation is obvious, never that the change is wanted — and the comment may be someone else's, so the user decides whether it lands at all. One OK settles the whole list; it applies nothing. Anything the user pulls out of it — _"all but the third"_ — turns HITL and gets grilled with the rest, so the OK is worth waiting for even though no code moves. ## 4. Grill the HITL ones Run a `/grill-me` session over the HITL comments. Three things are specific to this one: - **The shared understanding is per comment, not per batch.** Settle one and record what was agreed before opening the next. - **Two comments that are the same decision** in two places: merge them and say so. - **A comment that runs past one exchange**: park it and keep going with the rest, so one long argument holds nothing else hostage. A comment can settle as **leave the code as it is**. If the discussion produced a reason that is invisible from the code — a constraint, a trade-off already tried — offer to leave it as a code comment there, and it joins the batch like any other change. The user decides; a codebase padded with defensive comments is worse than the debate repeating. ## 5. Hand back the agreed batch With every comment settled, write the batch out: one line per comment — AFK and grilled alike, the ones settled as leave-it-alone included — in the same shape as the map in step 3. This is the only place the agreed work exists in writing, and it is what the user reads to decide. Then ask, proposing nothing: **what happens to this now?** ## 6. Wait Their answer is the whole of what happens next — implementing it here, writing it up, dropping it. Nothing moves before it arrives, and whatever it turns out to be, closing every comment it satisfies is part of doing it: a change that lands and leaves its comment standing reads as work still owed.
From javierponferradalopez/skills, MIT.
More from javierponferradalopez/skills
- 0.565score
branch
javierponferradalopez
Create a git branch off the current one, named `<task-code>-<type>-<description>`.
SkillAFreshness: AClaude Code +9 - 0.561score
implement
javierponferradalopez
Build one issue through three fresh-eyes phases — build, harden, gaps — write its docs with you, and report only what needs your eye; stops with the tree dirty.
SkillAFreshness: AClaude Code +9 - 0.558score
done
javierponferradalopez
Push the branch and close the issue it finishes — one approval gate, then push, then close.
SkillAFreshness: AClaude Code +9 - 0.553score
github-pr
javierponferradalopez
Prepare and create a high-quality GitHub Pull Request for the current branch
SkillAFreshness: AClaude Code +9 - 0.539score
domain-modeling
javierponferradalopez
Build and sharpen a project's domain model.
SkillAFreshness: AClaude Code +9 - 0.539score
tdd
javierponferradalopez
Test-driven development with red-green-refactor loop.
SkillAFreshness: AClaude Code +9 - 0.539score
improve-codebase-architecture
javierponferradalopez
Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/.
SkillAFreshness: AClaude Code +9 - 0.529score
diagnose
javierponferradalopez
Diagnosis loop for hard bugs and performance regressions.
SkillBFreshness: BClaude Code +9 - 0.521score
prune-comments
javierponferradalopez
Prune comments — every comment in scope faces one test, and the ones carrying no why go; survivors compress to one line.
SkillBFreshness: BClaude Code +9 - 0.519score
handoff-grill
javierponferradalopez
Pause a grill-me session into a resumable handoff that preserves the open branches of the decision tree, not just the closed decisions.
SkillBFreshness: BClaude Code +9 - 0.496score
to-tickets
javierponferradalopez
Break a plan, spec, or the current conversation into a set of tracer-bullet tickets, each declaring its blocking edges, published to the configured tracker —…
SkillBFreshness: BClaude Code +9 - 0.496score
to-spec
javierponferradalopez
Turn the current conversation into a spec and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 1
- Last commit (this folder)
- Aug 17, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/javierponferradalopez/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from javierponferradalopez
- 0.565score
branch
javierponferradalopez
Create a git branch off the current one, named `<task-code>-<type>-<description>`.
SkillAFreshness: AClaude Code +9 - 0.561score
implement
javierponferradalopez
Build one issue through three fresh-eyes phases — build, harden, gaps — write its docs with you, and report only what needs your eye; stops with the tree dirty.
SkillAFreshness: AClaude Code +9 - 0.558score
done
javierponferradalopez
Push the branch and close the issue it finishes — one approval gate, then push, then close.
SkillAFreshness: AClaude Code +9 - 0.553score
github-pr
javierponferradalopez
Prepare and create a high-quality GitHub Pull Request for the current branch
SkillAFreshness: AClaude Code +9 - 0.539score
tdd
javierponferradalopez
Test-driven development with red-green-refactor loop.
SkillAFreshness: AClaude Code +9 - 0.539score
domain-modeling
javierponferradalopez
Build and sharpen a project's domain model.
SkillAFreshness: AClaude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get grill-me-comments running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
grill-me-comments skill. Agent Mart. https://agentmart.dev/skills/javierponferradalopez-grill-me-comments (last updated 2026-10-24)