菜单
Builds a runnable skill from a plan produced by agent-plan. Writes SKILL.md and all named ref files to .claude/skills/<name>/. Use when handed a plan path, or when the user says "build this", "generate the skill", or "create the 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
| name | agent-build |
| description | Builds a runnable skill from a plan produced by agent-plan. Writes SKILL.md and all named ref files to .claude/skills/<name>/. Use when handed a plan path, or when the user says "build this", "generate the skill", or "create the agent". |
| allowedTools | ["Read","Write","Bash","TodoWrite"] |
Invoke: /agent-build <plan-path> or handed a plan path directly from agent-plan.
If no plan path is given, list plans/ and ask the user to confirm which plan to build before proceeding.
| Name | Format | Source |
|---|---|---|
| plan | markdown per agent-plan plan-template | plans/<name>.md |
| Name | Format | Destination |
|---|---|---|
| SKILL.md | skill file per refs/skill-template.md | .claude/skills/<name>/SKILL.md |
| refs/* | one file per entry in plan §6 | .claude/skills/<name>/refs/<filename> |
Emit Step X/5 — <title> at the start of each step, unconditionally.
Step 1/5 — Load and validate plan
Read the plan file. Check every required section is present and non-empty:
| Section | Required content |
|---|---|
| §1 Skill identity | YAML block with name and description |
| §2 Trigger conditions | At least one trigger |
| §3 Persona | All seven axes present |
| §4 Inputs and outputs | At least one input and one output |
| §5 Workflow | Diagram and protocol both present |
If any required section is missing or empty, stop. Emit:
Build halted: §<N> (<title>) is missing or empty.
Do not write any files.
Step 2/5 — Scaffold directories
Extract <name> from plan §1 YAML name field. Check for output_dir in plan §1:
output_dir is set (Desktop or Custom install): skill_root = <output_dir>output_dir is absent (Global install): skill_root = .claude/skills/<name>Run via Bash:
mkdir -p <skill_root>/refs
Do not prompt. If the directory already exists, continue.
Step 3/5 — Write SKILL.md
Map plan sections to SKILL.md using this table:
| SKILL.md element | Source |
|---|---|
Frontmatter name | Plan §1 YAML name |
Frontmatter description | Plan §1 YAML description |
Frontmatter model | claude-sonnet-4-6 unless plan §1 specifies otherwise |
## Usage / Invoke line | Slash command from plan §2 trigger conditions |
## Usage / trigger list | All natural-language and context triggers from plan §2 |
## Inputs table | Plan §4 Inputs |
## Outputs table | Plan §4 Outputs |
## Persona section | Plan §3 — all seven axes verbatim |
## Step-by-step protocol | Plan §5 Protocol — numbered steps; tighten language per language.md |
## References bullets | Plan §6 — one bullet per ref, filename.md — one-line purpose |
Follow refs/skill-template.md for section order and formatting.
Apply refs/principles.md quality gates before writing.
Write to <skill_root>/SKILL.md. Do not prompt.
Step 4/5 — Write ref files
For each entry in plan §6, write one file to <skill_root>/refs/<filename>.
Frontmatter format (required on every ref):
---
name: <Title Case name>
description: <one-line purpose, copied from plan §6 bullet>
type: reference
---
Content generation rules:
| Ref type | How to generate |
|---|---|
| Domain standard (WCAG, Nielsen's heuristics, RFC, accessibility specs) | Generate from domain knowledge. Include all canonical items — never stub. |
| Persona-derived (refusal phrasings, tone guides, decision heuristics) | Derive from plan §3 axes 4, 6, and 7. |
| Workflow-derived (checklists, taxonomies, state lists) | Derive from plan §5 workflow steps and plan §3 axis 3 (Knowledge & expertise). |
| Proprietary or custom process | Derive what can be inferred from the plan. Mark gaps with [USER: fill in <specific item>] — never leave silent stubs. |
Minimum content per ref:
Every ref must be self-contained: each rule understandable without reading another file.
Step 5/5 — Verify and emit
For each file written, read it back and check it against refs/principles.md. If any gate fails, rewrite and re-check.
Emit a build report:
Build complete: <skill_root>/
SKILL.md <N> lines
refs/<file1>.md <N> lines
refs/<file2>.md <N> lines
...
Then hand off: "Ready for /agent-evaluate."
Apply refs/principles.md before writing each file. Apply refs/language.md to all prose output.
refs/skill-template.md — required SKILL.md output structurerefs/principles.md — quality rules applied before writing each output filerefs/language.md — voice and terminology for all output proseSenior product designer that audits UI screens and PRDs for usability flaws, missing error states, edge-case gaps, and accessibility issues. Use when the user says "review this design", "audit this UI", "critique this flow", or invokes /design-review with a screenshot and a PRD.
Fifth subagent in the agent-audit pipeline. Reads evals-[n].json and grading.json from the run dir, aggregates timing across all evals, computes pass rate and token stats, and writes timing.json and benchmark.json. Handles missing or partial data gracefully — writes what it can and flags gaps. Use when agent-audit hands off "compute the benchmark".
Second subagent in the agent-audit pipeline. Grades every assertion in evals-[n].json against actual_output using LLM judgment for semantic checks and tool calls for mechanical checks. Writes grading.json with per-assertion pass/fail and evidence, and updates evals-[n].json with a verdict per case. Use when agent-audit hands off "grade the evals".
Third subagent in the agent-audit pipeline (comprehensive mode only). Checks the environment for agentlinter and agnix, installs missing tools via npm if possible, runs each linter independently, then runs an LLM safety check against the audit registry. Each check is independent — a missing or failing tool is flagged and skipped, never blocks the others. Writes audit-[n].json to the run dir. Use when agent-audit hands off "run lint and safety check".
Fourth subagent in the agent-audit pipeline (comprehensive mode only). Measures the target skill's description trigger rate, iterates to improve it, and writes the winning description back to SKILL.md. Optimised for token efficiency: train iterations use 1 run per query; validation and baseline use 3. Stops early at 90% train pass rate. Appends optimiser results to audit-[n].json. Use when agent-audit hands off "run the description optimiser".
Orchestrator for the agent-audit pipeline. Prepares the run directory, asks the user which checks to run via a labelled checklist, dispatches specialist subagents in the correct dependency order, and produces a final report. Use when the user says "audit this skill", "run agent-audit", "check this agent", or agent-evaluate hands off "run audit".