菜单
Use to add validation criteria to a spec or plan. Triggers include "define validation", "how will we validate", "what should the AI check", and "add validation to this".
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
Use when checking whether field-guide reference pages still match current code and tests. Reads pages plus fg-evidence anchors, reports or fixes semantic drift, and can use sub-agents for parallel audit batches. Triggers include "resolve drift", "semantic drift", "check accuracy", "verify the reference against code", and "audit the wiki".
Use after ingesting or editing field-guide pages to attach living code/test anchors. Adds or refreshes fg-evidence metadata without trying to prove prose accuracy. Triggers include "update evidence", "wire evidence", "add evidence anchors", and "connect the wiki to code".
Use when extracting knowledge from .lore/ artifacts into the reference wiki. Converts work artifacts (retros, specs, plans, research, notes) into structured wiki pages grouped by knowledge type. Reads Markdown or HTML sources and writes Markdown by default. Triggers include "ingest this", "extract knowledge from", "add this to the wiki", "update the reference from", and "populate the field guide".
Use when setting up field-guide on a project for the first time, or to refresh the scheduled lint job after it expires. Triggers include "init field-guide", "set up field-guide", "refresh field-guide lint schedule", and "bootstrap the wiki".
This skill is used when the project wiki needs a health check. Triggers include 'lint the wiki', 'check wiki health', 'run field-guide lint', 'are there stale wiki pages', 'find orphaned pages', 'check for contradictions in the wiki'.
Use when asking questions about the project wiki, querying accumulated knowledge, searching project decisions or lessons. Triggers include "query the wiki", "what does the wiki say about", "search project knowledge", and natural language questions directed at the field-guide wiki.
| name | define-validation |
| description | Use to add validation criteria to a spec or plan. Triggers include "define validation", "how will we validate", "what should the AI check", and "add validation to this". |
Validation answers: "How will the AI know this is done?"
It must be behavioral and actionable — something the AI can actually run or observe. "Run the CLI with these args and confirm the output contains X." "Use Playwright to confirm clicking the button triggers Y." "Run the test suite and check it passes." Not "verify the UX feels right."
Add a validation section to an existing spec or plan. If none exists, save standalone to .lore/work/validation/[topic].md. Load ${CLAUDE_PLUGIN_ROOT}/shared/frontmatter-schema.md for the frontmatter fields before writing.
When saving standalone, write the body in Markdown per the "Body Format" section of ${CLAUDE_PLUGIN_ROOT}/shared/frontmatter-schema.md. Reach for embedded inline HTML only when a visual carries meaning prose can't, such as a diagram separating automated from manual steps.
Structural assertions ("verify this function appears only once") don't count. If a regression must not recur, write a test that fails when the behavior regresses, not when lines move.
