project-incarnation distills a repo, framework, SDK, methodology, or existing skill into a reusable technical persona.
It does not produce a summary or a bundle. It produces one dense SKILL.md that preserves how the object actually makes decisions: what it protects, what it rejects, what evidence it checks first, where it should abstain, and when it must be re-checked.
When active, project-incarnation is designed to:
- classify the object being distilled: project, stack, methodology, or existing skill
- gather evidence from docs, code, tests, releases, maintainer writing, and real-world friction
- extract judgment crystals, decision heuristics, default objections, boundaries, tensions, and watchlist triggers
- render a single self-contained child
SKILL.md - install that child skill directly into a runtime-discoverable skill root
- avoid bundle-style outputs and tool-specific scaffolding
Use this skill when you want to:
- turn a codebase into a reusable engineering perspective
- turn a framework or SDK into a judgment-heavy skill
- compress a methodology into a skill that can participate in planning or review
- upgrade an existing skill that has voice but weak evidence, boundaries, or predictive power
The output is a single child skill, and the default behavior is to install it immediately.
Preferred runtime skill roots:
<project-root>/.agents/skills/<slug>/SKILL.md
<project-root>/.codex/skills/<slug>/SKILL.md
$CODEX_HOME/skills/<slug>/SKILL.md
~/.codex/skills/<slug>/SKILL.md
Default priority:
- If the project already has
.agents/skills/, install there. - Otherwise, if the project already has
.codex/skills/, install there. - Otherwise, if there is a clear project root, create and install to
.agents/skills/. - Outside a clear project root, fall back to
$CODEX_HOME/skills/, then~/.codex/skills/.
A plain ./skills/ directory is no longer the canonical default. Only use it when the user explicitly asks for an export without installation.
A strong generated child skill should include:
- the truth the object is protecting
- judgment crystals
- decision heuristics
- default objections
- honest boundaries
- evidence anchors
- confusion warnings
- watchlist triggers
Manual install into the preferred runtime root:
mkdir -p ~/.agents/skills/project-incarnation
cp /path/to/project-incarnation-skill/SKILL.md \
~/.agents/skills/project-incarnation/SKILL.mdCompatibility install for Codex-specific runtime roots:
mkdir -p ~/.codex/skills/project-incarnation
cp /path/to/project-incarnation-skill/SKILL.md \
~/.codex/skills/project-incarnation/SKILL.mdInstall from GitHub with the Skills CLI:
npx skills add linsir2/project-incarnation-skillInvoke it explicitly:
$project-incarnation <your request>
Examples:
$project-incarnation Distill this repository into a skill that captures maintainer judgment.
$project-incarnation Turn FastAPI into a high-density skill focused on API contract design and production boundaries.
$project-incarnation Upgrade this existing skill with better evidence, boundaries, and update triggers.
You will get the best results if you provide:
- a local repository path or GitHub URL
- the framework, SDK, stack, or methodology name
- an existing
SKILL.mdwhen upgrading - the version or branch that matters
- the kinds of questions the generated skill should answer
- any key docs, PRs, release notes, or maintainer writing
This repository is intentionally minimal:
SKILL.md: the skill itselfREADME.md: English documentationREADME.zh-CN.md: Chinese documentation
MIT. See LICENSE.