菜单
Author and verify engineering rules as a single HTML contract. Use when the user runs /html-as-contract to review, accept, or suggest rules, or /html-as-contract check to scan code and regenerate the compliance report tab. Callable by a human or by another agent.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Read the current project's current and prior Claude Code or Codex sessions, extract every prompt, score five effort metrics 0-100 with reasons, estimate average tokens and time per session, and render a self-contained light-theme HTML report. Use when the user runs /effort-map or asks how much effort, thought, bug-fixing, UX tuning, architecture care, or copy-paste slop went into a project.
Find bugs that were encountered and solved in prior Claude Code or Codex sessions for the current project, then generate a self-contained searchable light-theme HTML bug report. Use when the user asks for a bug catalog, solved-problem history, prior-session bug report, recurring failure inventory, or a BUG to Solution report for the current repository.
Analyze a prompt with Edward de Bono's Six Thinking Hats and produce a polished, self-contained HTML report. Use when the user invokes /hats or $hats, requests Six Thinking Hats analysis, wants a decision reviewed from factual, emotional, cautious, optimistic, creative, and process perspectives, or asks for an illustrated HTML decision report.
Reads every Claude Code session in the user's config directory and renders a self-contained light-theme website with a GitHub-style contribution grid, visual metrics, and 5-10 written insights about the user's agent coding habits and what they are building. Use when the user runs /habit-tracker or asks about their coding habits, activity streak, contribution graph, what they have been building, or a recap of their Claude Code usage.
Rewrite poor git commit messages into a clear conventional-commit history. Reads the last N commits with their diffs, proposes a clean conventional message for each by reading what the code actually changed, then opens a light-theme review UI where the user approves all or each message individually before the history is rewritten. Use when the user runs /fix-git-history or asks to clean up, rewrite, normalize, or fix commit messages or git history into a conventional-commit format.
Audits a running web page for WCAG accessibility issues, captures a screenshot and annotates it with a numbered box over every problem, then renders a self-contained light-theme website with the score, the annotated screenshot, and a fix for each finding. Use when the user runs /accessibility-auditor or asks to audit, check, or report a site's accessibility, WCAG compliance, alt text, color contrast, labels, or screen-reader support.
| name | html-as-contract |
| description | Author and verify engineering rules as a single HTML contract. Use when the user runs /html-as-contract to review, accept, or suggest rules, or /html-as-contract check to scan code and regenerate the compliance report tab. Callable by a human or by another agent. |
A single contract.html holds engineering rules (tab 1) and a live compliance
report (tab 2). It is dual-reader: a human reads the rendered light-theme page,
an agent reads the same markup and writes the report back into it.
Dispatch on the argument after the command.
contract.html in the current working directory, or the path
the user names.template.html.#rules (tab 1) is human-authored. In every mode except review/suggest it is
read-only. Never rewrite it during a check.#compliance-report (tab 2) is agent-owned and replaced wholesale on check.#agent-protocol are fixed. Do not touch them./html-as-contractcontract.html is missing, copy template.html to contract.html.#rules and present them.#rules only.#compliance-report./html-as-contract suggestcontract.html exists; scaffold from template.html if missing.#rules./html-as-contract checkcontract.html and follow the embedded #agent-protocol..rule in #rules, scan the scope (default project root; the sample
scans sample/src).pass / fail / unknown, honoring data-check: static is a
deterministic pattern match, semantic is judgment and must state confidence.#compliance-report: the summary stats, a
.meta line with generated-at and scope, and one .report-item per rule with
data-rule-id and data-status, citing each violation as file:line.#rules. This mode is non-interactive and safe for a human or
an agent to call.Each rule is:
<article class="rule" data-rule-id="..." data-category="testing|design|code"
data-severity="error|warning|info" data-check="static|semantic">
<h3>title <span class="badge error">error</span> <span class="badge check">static</span></h3>
<p class="statement">the rule</p>
<p class="rationale">why it matters</p>
<div class="pair">
<div class="col"><p class="label good-label">good</p><pre class="good">...</pre></div>
<div class="col"><p class="label bad-label">bad</p><pre class="bad">...</pre></div>
</div>
</article>
Rules are grouped under <h2 class="category"> headings per category.
<div class="summary">
<div class="stat pass"><div class="num">N</div><div class="lbl">passed</div></div>
<div class="stat fail"><div class="num">N</div><div class="lbl">failed</div></div>
<div class="stat unknown"><div class="num">N</div><div class="lbl">unknown</div></div>
</div>
<p class="meta">Generated <iso-timestamp> · scope <path></p>
<div class="report-item" data-rule-id="..." data-status="pass|fail|unknown">
<h4>title <span class="status-tag">fail</span></h4>
<div class="violation"><code>file:line</code> — short reason</div>
</div>