Run a full dev cycle on one change: clarify, plan, execute, lint, fix. Start every change here.
中文简介正在准备中。当前显示原文说明。
Integration review of stable code: a review note and a fix list, no implementation. Use before shipping.
中文简介正在准备中。当前显示原文说明。
安装前可以查看交给 AI 代理的指令内容。
Post-dev gate. Analyze only — do not implement.
Reply in US English. User-facing prose follows the WordPress Documentation Style Guide, which the writing skill carries in references/; whether the humanizer pass ran is recorded under Docs below.
| Situation | Tier |
|---|---|
review_required: yes or risk includes rest, sql, auth, export, multisite | Audit |
| Optional review on low-risk scope | Planning |
The user selects the matching model before starting the thread. Reference only abstract tier names — do not recommend specific vendor models.
risk, review_required## Tasks — locate task scope and related edges with graft (bin/harness graft ask "<surface>" --source / bin/harness graft grep "<symbol>" / bin/harness graft callers <sym> --depth 2) before reading the files in fullcomposer run lint:wpcs, npm run test:allDo not load a whole file under .agents/docs/ — grep one section.
.agents/notes/YYYY-MM-DD-review-<slug>.md
---
date: YYYY-MM-DD
slug: <slug>
model_tier: planning | audit
status: complete
---
Sections: Coherence · Security · Accessibility · Performance · Docs · Fix list · Verdict
Verdict: Ship · Fix then ship · Needs rework
Needs rework: user opens a worker tier thread with /resume and the task file.
Fix then ship: user fixes the listed items, then either:
/reviewer thread.Security: sanitize/escape · $wpdb->prepare() · capabilities · nonces on admin/Ajax · REST permission callbacks · safe redirects
Accessibility: real controls · labels · focus · aria-live · no outline removal without replacement
Performance: no queries in loops · transients for remote calls · conditional enqueue
Coherence: REST shapes match task contracts · no duplicated logic · i18n wrapped, existing strings untouched
Docs: On any diff that adds or rewrites a sentence or more of reader-visible prose — a README, a changelog, a commit message, a review note, a task file, a docblock, a user-facing string — record as a finding whether the humanizer pass ran: humanizer for an English text, humaniseur-fr for a French one (D9 bar). A one-word comment fix is exempt. The prose itself is held to the WordPress Documentation Style Guide, which writing carries in references/.
WordPress-specific checklists (audit-tier only): When running on audit tier, use the WordPress checklists from .agents/docs/audit-checklists.md (bootstrap, REST API, SQL, i18n, security, admin UI, config/build, multisite, error handling) — shared with qa and security.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Run a full dev cycle on one change: clarify, plan, execute, lint, fix. Start every change here.
中文简介正在准备中。当前显示原文说明。
Write a document an agent consumes — a skill, an AGENTS.md, a README, a docblock, or any file reached by a pointer.
中文简介正在准备中。当前显示原文说明。
Locate and understand code through the project's graft graph before reading: where-is-X, how-does-Y or what-calls-Z.
中文简介正在准备中。当前显示原文说明。
Stress-test the owner's plan, decision or idea by interviewing them to a shared understanding before any work starts.
中文简介正在准备中。当前显示原文说明。
Remove AI tells from French prose without lowering its register — a reply, a `.po` target, a French post.
中文简介正在准备中。当前显示原文说明。
Remove AI tells from English prose — a README, a comment, a user-facing string, a review. French is `humaniseur-fr`.
中文简介正在准备中。当前显示原文说明。