pr-review-checklist
Use when reviewing code/manifests for merge, approving a PR, or requesting changes. Applies scope, secrets, seam, load-bearing-assumption and AI dark-pattern checks, then each touched stack's own review checks.
- License 许可
- MIT
- 所属仓库
- pixeloven/crew
- 最近更新
- 2026-11-07
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- pr-review-checklist
- tier
- concept
- requires
- []
- description
- Use when reviewing code/manifests for merge, approving a PR, or requesting changes. Applies scope, secrets, seam, load-bearing-assumption and AI dark-pattern checks, then each touched stack's own review checks.
- expects-local
- [ "platform-conventions", "protected-seams" ]
03 SKILL.md 正文
Every review opens with a one-sentence summary judgment (Pass / Pass with required changes / Block), then lists findings by category: Required, Recommended, Note, and Escalations for findings that would amend the accepted ask or repeat one theme across rounds (per delta-handling).
When the review is posted to the PR, it starts with the agent marker 🤖 **Agent**: — on the review body and on every inline comment — so the author can tell agent findings from human ones. Never post an agent review unmarked, and never add the marker to text a human wrote.
Universal checks (all PRs)
- [ ] PR scope matches stated purpose — no unrelated changes bundled
- [ ] Load-bearing assumptions listed, and each one falsified against source. This is the highest-yield check in the list: the common defect is a claim the author asserted but never ground-truthed. If the PR names no assumptions, that is itself a finding — every non-trivial change rests on something.
- [ ] No secrets hardcoded in any changed file (tokens, passwords, API keys)
- [ ] The project's protected-seams registry checked, if it defines one — per
seam-detection - [ ] Commit message follows
<type>(<scope>): <subject>format
Code
- [ ] The project's lint, format and test gates pass (its
AGENTS.mdor the stack's conventions skill names them) - [ ] MCP tools: tool descriptions are precise (agents use them for selection)
- [ ] No parallel implementation of a capability that exists in MCP — CLI is a thin wrapper
Stack checks
For each stack the diff touches (Kubernetes manifests, Terraform, Ansible, CI workflows, a language), apply that stack's conventions skill's review checks. A stack with no conventions skill installed gets the universal checks above plus the project's platform-conventions; say in the review that no stack checks were applied.
AI behaviour and narrative checks
Apply these when the diff ships behaviour an AI system presents to people (an assistant's wording, a recommendation, an automated decision), or when it is itself a brief, spec, plan or research write-up that argues for building something. Name the pattern in the finding.
When the diff ships AI behaviour, check it does none of these:
- [ ] Anthropomorphic manipulation: the system voices feelings (hurt, disappointment, loyalty) to make the user feel guilt, attachment or obligation toward it
- [ ] Manufactured dependency: the assistance wears down what the user can do unaided, for example by hiding the steps it took or the means to do the task by hand
- [ ] Simulated understanding: wording claims comprehension of context, emotion or intent the system does not process, so users trust it beyond what it does
- [ ] Undisclosed AI decisions: a consequential outcome (eligibility, price, ranking, moderation) is decided by a model without telling the person it affects
When the diff is a brief, spec, plan or research write-up, check the argument against its evidence:
- [ ] Smoothed-arc composite: one tidy user story stands in for research that showed several paths that do not converge
- [ ] Manufactured-tension brief: the problem is sized to fit the solution already chosen, rather than to what the evidence shows
- [ ] Story as evidence: a vivid anecdote carries a decision the cited data does not support, or the data is gathered after the decision
Adapted from Design with Intent's anti-pattern catalog (github.com/ghaida/intent, CC0 1.0).
Seam check summary
After completing surface-specific checks, explicitly state:
- Which (if any) registry seams were touched
- Whether each crossing was flagged by the author
- Whether human sign-off is needed before merge
04 同仓库更多 Skills
orchard-gh-runs
Use when reading a failed GitHub Actions run with the gh CLI in this project, before re-running it or reporting a CI failure, so the failing step's log is read instead of the whole run.
orchard-gh
Hand-edited after promotion; the learned marker was dropped, so Doctor and check_skills --local can no longer see it.
demo-topology
Use when exercising the inventory contract fixture so that this fixture description comfortably clears the schema-v2 minimum length.
helper-jq
Use when exercising the inventory contract fixture so that this fixture description comfortably clears the schema-v2 minimum length.
user-notes
Use when exercising the inventory contract fixture so that this fixture description comfortably clears the schema-v2 minimum length.
toolkit-gh
Use when exercising the inventory contract fixture so that this fixture description comfortably clears the schema-v2 minimum length.