菜单
Autonomous codebase health scan for vocabulary drift, naming convention violations, structural anti-patterns, documentation staleness, and test health. Produces prioritized findings as Linear issues. Use for health checks, scheduled scans, or after heavy unreviewed agent work.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/outfitter-dev/trails --skill clark-survey命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | clark-survey |
| description | Autonomous codebase health scan for vocabulary drift, naming convention violations, structural anti-patterns, documentation staleness, and test health. Produces prioritized findings as Linear issues. Use for health checks, scheduled scans, or after heavy unreviewed agent work. |
| context | fork |
| agent | clark |
Autonomous scan. Broad, pattern-oriented. Not reviewing any single PR or milestone, but looking across the codebase for accumulating drift, emerging patterns, and opportunities to tighten.
Designed to run on a schedule (nightly, weekly) or on-demand when a general health check is needed. Produces actionable findings as Linear issues.
Search the codebase for terms that do not belong. Same vocabulary check as calibrate, but applied to the entire codebase rather than a specific changeset.
# Example searches to run (adapt as needed)
rg -i "handler|middleware|endpoint|controller" --type ts -g '!node_modules'
rg -i "registry|collection|manifest" --type ts -g '!node_modules' -g '!package.json'
rg -i '"serve"|"mount"|"wire"' --type ts -g
Not every match is a violation. Context matters. But every match is worth examining.
Look at recent exports and public API additions. Do they follow ADR-0001?
rg "^export " --type ts -g '!node_modules' -g '!*.test.*' -g '!*.spec.*'
Check for: factories without create*, derivations without derive*, test helpers without test*, validators without validate*.
Look for code smells that indicate architectural drift:
Request, Response, McpSession in trail filesthrow statements in implementation code (should be Result.err())console.log, console.error in non-surface code.run() calls: should be ctx.compose()rg "throw " --type ts -g '*/trails/*' -g '!*.test.*'
rg "console\." --type ts -g '!node_modules' -g '!*.test.*'
rg "\.run\(" --type ts -g '*/trails/*'
Compare the docs against the code:
lexicon.md cover all exported terms?architecture.md match the actual package structure?horizons.md current? Have any horizon items shipped without being moved to "shipped"?bun run test pass cleanly?trails warden report violations?bun run test
bun run typecheck
trails warden
trails survey brief
Look for quantitative signals:
## Survey: [date]
### Critical (file immediately)
- [finding]: [specific files/lines, what is wrong, recommended fix]
### Important (file this week)
- [finding]: [description, impact, recommendation]
### Minor (file for backlog)
- [finding]: [description, low urgency but worth tracking]
### Healthy
- [areas that look good, briefly]
### Trends
- [patterns across multiple findings]
- [things getting better or worse over time]
For each finding that warrants an issue:
**Title:** [Clark: brief description]
**Labels:** clark-survey, [area: vocabulary | architecture | naming | testing | docs]
**Priority:** [urgent | high | medium | low]
**Description:**
[What was found]
[Where it was found (files, lines)]
[Why it matters (which principle/convention)]
[Recommended fix]
If a previous survey exists at .trails/clark/survey-latest.md, read it before running. Note:
After the survey, write the current findings to .trails/clark/survey-latest.md so the next run has context.
docs/lexicon.md — lexicon to enforcedocs/adr/001-naming-conventions.md — naming conventions to checkdocs/architecture.md — structural expectationsAGENTS.md — current conventions.trails/clark/survey-latest.md — previous survey results (if exists)Run a Trails Regrade migration loop with dry-run discovery, occurrence judgment, safe apply, verification, ledger capture, and per-cycle resume.
Trails prose craft and lexicon style. Use when writing or reviewing docs, ADRs, examples, release notes, agent prompts, comments, PR descriptions, or issue language for rhythm, clarity, and vocabulary precision.
Run a Trails Regrade migration loop with dry-run discovery, occurrence judgment, safe apply, verification, ledger capture, and per-cycle resume.