ship-review
Skill0 starsChanged 41 days agoMITView on GitHub
Turn code review findings that already exist in context into a posted GitHub PR review written in the user's voice. Use when findings from a review skill are in context and the user says \"post this as a review\", \"comment these on the PR\", \"put this on github\".
1 other project uses a version of this.
Adopted by
Most recently taken up by
- HarambeeAI/bayes-shiptheir own version, first seen Apr 15, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- HarambeeAI/bayes-ship★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- grill-meSkill
- multiphase-pr-review-asyncSkill
- multiphase-pr-reviewSkill
- pr-splitSkill
- publish-prSkill
- take-feedbackSkill
- thermo-nuclear-code-quality-reviewSkill
- ticketedSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
If there are no review findings in context, stop and say so.
- Distinctive: line 17
Did the PR introduce it, or inherit it?: Read the pre-image. This is the single biggest severity driver.
- Distinctive: line 18
Trace the runtime claim to its end: If you claim a 500, find the handler.
- Very specific: line 27
Use AskUserQuestion. Batch 2-3 findings per round, grouped by what couples them (same subsystem, same root cause).
- Distinctive: line 29
Always offer "Drop it" as an option, with the honest case for dropping.
- Distinctive: line 52
COMMENT: when something genuinely needs an answer before merge.
- Very specific: line 53
REQUEST_CHANGES: only for damaging issues: data loss, security, a new defect with real consequences that this PR introduces.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Ship Review
Takes findings that already exist and turns them into a posted review. Does not produce findings.
If there are no review findings in context, stop and say so. Point at the review skills that produce them. Do not audit the diff yourself to fill the gap.
1. Re-verify before presenting anything
Findings from a review pass are candidates, not conclusions. Before grilling, check each one:
- Does the repo actually support the complaint? A generic rule can be a non-issue locally. Check neighbours before flagging file size, naming, layering, test count, DI style. If the norm contradicts the finding, drop it and say plainly that you are dropping it and why.
- Did the PR introduce it, or inherit it? Read the pre-image. This is the single biggest severity driver.
- Trace the runtime claim to its end. If you claim a 500, find the handler. If you claim broken access, find the check. Unverified consequences do not get posted.
- Look once more for what the review missed, especially anything on the theme the user cares about. New findings are welcome here; unverified ones are not.
Report corrections to your own earlier findings flatly, no hedging or apology.
2. Grill for severity and inclusion
The user decides what ships. You supply context and a recommendation.
- Use
AskUserQuestion. Batch 2-3 findings per round, grouped by what couples them (same subsystem, same root cause). Do not fire one question per finding. - Recommended option first, labelled
(Recommended). - Always offer "Drop it" as an option, with the honest case for dropping.
- Give the context they may not have: what the mechanism is, what the blast radius is, whether it is pre-existing. Two or three sentences.
- Offer severity shapes, not just yes/no: ask with a concrete fix / flag and ask if intentional / note without any ask / drop.
- Nits get one batched question at the end, multi-select.
- Answer their questions straight. If they ask whether something buys anything and it does not, say it does not.
3. Draft in the user's voice
- Short. One to three sentences. Longer only to propose a concrete restructure.
- Ask, do not assert. "Intentional?" "Align?" "Or whats your thoughts?" "...right?" A question invites the author's context; a statement invites defending.
- No emdashes. Ever. Commas, periods, or parens.
- Casual and lowercase-ish. Contractions without apostrophes read natural here: doesnt, thats, havent, isnt. "Hmm," to open a doubt. abit, mby, rn.
nit:prefix for cosmetics. "not blocking" / "fine as a follow-up" for optional asks.- Pair every complaint with a concrete alternative or a ```suggestion block. Never leave a complaint with no exit.
- Show real uncertainty where it exists. "Dont know the legacy data well enough to say." "Dunno X well." Never fake it, and never hedge a verified fact.
- Say when something is pre-existing. It changes how the author hears it.
- Direct about real problems ("This is spaghetti.") but never dismissive of the person. The fix follows immediately.
- Sparing emoji: 👍 🤔 😄
- Credit what the PR got right, in the summary body, specifically. Vague praise reads as filler.
4. Verdict bar
- APPROVE with comments is the default, including for substantive optional asks the author can take or leave.
- COMMENT when something genuinely needs an answer before merge.
- REQUEST_CHANGES only for damaging issues: data loss, security, a new defect with real consequences that this PR introduces. Inherited warts never reach this bar.
Summary body follows the same voice: two or three lines. Specific credit, the verdict, the one or two things worth looking at, "rest is nits."
5. Approve, then post
Present every comment verbatim with its path:line anchor, plus the summary body and verdict. Then ask for the go via AskUserQuestion with a "let me edit first" option. Never post unasked.
Anchor lines must exist in the diff or GitHub silently drops the comment. Read the file with line numbers to pin anchors; do not count by eye.
gh pr list --head "$(git branch --show-current)" --state all --json number,url
# write review.json to the scratchpad: {event, body, comments:[{path, line, side:"RIGHT", body}]}
gh api /repos/OWNER/REPO/pulls/N/reviews -X POST --input review.json -q '{state, url: .html_url}'Then read back what actually landed, and report any anchor that went missing:
gh api "/repos/OWNER/REPO/pulls/N/comments?per_page=100" -q '.[] | select(.user.login=="LOGIN") | "\(.path | split("/") | last):\(.line)"'Close by telling the user what shipped and what got dropped, and flag anything worth a ticket rather than a PR fix.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

