| name | author-workflow-skill |
| description | Create a new project SKILL.md for a repeatable workflow. Use when user asks to add a skill for vulnerabilities, package upgrades, audits, supply-chain checks, or any named agent workflow. |
Author a workflow skill
When to use
User requests: "Add a SKILL.md for …", "create a skill to …", or describes a repeatable procedure worth attaching on demand.
Steps
- Glob existing
.cursor/skills/ and .agents/skills/ — match naming and frontmatter style.
- Read 1–2 existing SKILL.md files in the repo for tone and section layout.
- Draft skill with:
- YAML
name + description (third person; include trigger phrases)
- Formal-first body (optional dense header per formal/AppGenMathPhyLang.kernel.md; English in
(parentheses) where a symbol may confuse)
- When to use / skip — bullet triggers
- Steps — ordered, tool-friendly; one diagram if state machine/DAG
- Done when — verifiable exit criteria (commands/artifacts)
- references/english-*.md — full plain English; agent loads only if formal SoT is insufficient
- Write under
.agents/skills/<name>/SKILL.md (or project convention path).
- Optionally add a routing rule (
.mdc) telling the agent when to suggest attaching the skill.
- Meet QUALITY.md — no tool-name-only stubs.
Do not
- Paste long policy essays — skills are procedures, not style guides
- Dual-load formal + full English by default (token waste)
- Duplicate an existing skill — extend or link instead
- Commit secrets or environment-specific paths — use
{REPO_ROOT} placeholders
- Ship distill “evidence: turn N” archaeology as a skill
Validate
Ask user to invoke /skill-name on the next matching task; compare tool pattern to gold sessions (Read → research → Shell verify).
