pr
Skill5 starsChanged 10 days agoMITView on GitHub
The pull request body format, built for fast human review. Use when writing or rewriting a PR body.
In this repository57 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- architecture-satisfactionSkill
- autopilotSkill
- browser-testSkill
- chatter-scoutSkill
- closeoutSkill
- codebase-designSkill
- code-reviewSkill
- code-smells-auditSkill
- commitSkill
- dependency-auditSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- fix-ci-failuresSkill
- fix-github-issuesSkill
- fix-nightwatch-issuesSkill
- fix-posthog-issuesSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- harness-toolsSkill
- housekeeperSkill
- implementSkill
- implement-specSkill
- improve-codebase-architectureSkill
- laravel-herd-worktreesSkill
- panelSkill
- prototypeSkill
- prune-testsSkill
- releaseSkill
- repository-cleanupSkill
- researchSkill
- resolve-review-commentsSkill
- resolving-merge-conflictsSkill
- retroSkill
- run-afterSkill
- scratch-cleanupSkill
- scratch-statusSkill
- setupSkill
- sweepSkill
- tddSkill
- teachSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- triage-github-prSkill
- triageSkill
- ui-designSkill
- upgrade-planSkill
- verifySkill
- wait-whatSkill
- wayfinderSkill
- which-skillSkill
- wizardSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Use this template for the PR body.
- Distinctive: line 93
Place each visual next to the short text it supports.
- Distinctive: line 102
No evidence is a finding, not a blank: say what was not verified and why.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
PR
Use this template for the PR body. Skip preambles, keep prose brief, and use the project's domain language from GLOSSARY.md when it exists.
## Summary
<one visual: pseudocode, call tree, file tree, diagram, or diff-sketch>
## Evidence
- **Before:** <screenshot / output / failing test run>
**After:** <screenshot / output / passing test run>
## Merge Danger
**Door:** <one-way or two-way>
<optional: why>
**Blast radius:** <one word>
<optional: what could break on merge>Lines the caller needs (fixes #NNN, one line per closed ticket) go after Merge Danger, unchanged.
Summary
Pick the smallest view that makes the key point clear. Usually one; never all of them.
- Logic or an algorithm as pseudocode:
on(invoice paid)
if already reconciled
return
mark reconciled
dispatch ReceiptMailer- Runtime control flow as a call tree:
CheckoutController@store
PlaceOrder::handle
ReserveStock
ChargeCard
OrderPlaced event
SendConfirmation (queued)- File responsibility or a broad refactor as a shallow file tree:
app/
├── Actions/Billing/ # one class per billing operation
├── Models/Invoice.php # owns invoice state
└── Jobs/SyncStripe.php # pushes changes to Stripe- Interaction or data flow as Mermaid:
sequenceDiagram
participant User
participant App
participant Stripe
User->>App: submit checkout
App->>Stripe: create PaymentIntent
Stripe-->>App: webhook payment_intent.succeeded- A
diffsketch when the point is what changes inside a shape that already exists. Match the shape to the topic (call tree, file tree, state flow):
PlaceOrder::handle
ReserveStock
+ ApplyCoupon
ChargeCard
- SendConfirmation
+ OrderPlaced event
+ SendConfirmation (queued)- The whole block of code only when most of it is new, or when leaving context out would hide ownership or order.
Place each visual next to the short text it supports. Keep only the calls, files, states, and boundaries the reviewer needs.
Evidence
Concrete proof that the change works, as a before and after.
- Screenshots are best, when the change is visual and the environment can take them.
- Execution output is next: the exact test that failed and now passes (name it, and sketch its steps as pseudocode), or command output before and after.
No evidence is a finding, not a blank: say what was not verified and why.
Screenshots
When you open a PR, put every screenshot taken this session of what the PR builds into Evidence: from verify, from a browser-test run directory, or saved ad hoc. Leave out screenshots of other work and of failures you later fixed. Inspect each screenshot before uploading: never upload one that shows real customer or staff PII (names, emails, phone numbers) or secrets from a local database dump; re-shoot against fake/seeded records or redact first. No screenshots: skip this section and do not take any just for the PR; Evidence uses execution output. GitHub cannot attach a local file to a PR body, so upload them first. From inside the repo, run this skill's script:
bash <this skill's directory>/scripts/upload-screenshots.sh <pr-branch> <image>...It commits the images to the orphan pr-assets branch (the PR diff stays clean), pushes it, and prints one  line per image, linked through the branch path so it stays live until repository-cleanup prunes that PR's folder. Paste the lines under After:, or Before: for a screenshot of the old behaviour, each with a one-line caption of what it shows. If the push fails, keep the body, list the local paths under Evidence, and say the upload failed.
Merge Danger
Door. A two-way door is cheap to walk back: revert and redeploy. A one-way door is not: a destructive migration, a data backfill, a public API or contract change, an email sent, a queue payload shape that in-flight jobs still carry.
Blast radius. What the change could touch if it is wrong. Consider every consumer: other modules, queued jobs, API clients, layout, mobile, scheduled commands.
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.

