ai-project-skill-maker
ai-project-skill-maker is a meta-skill that generates durable project maintainer skills and instruction bundles for AI coding agents. Instead of producing ordinary project documentation, it creates project-specific maintainer guidance that future AI sessions can use to understand a repository, make changes in the right style, run the right checks, preserve project intent, and avoid casually editing generated or sensitive files.
Links
- Repository:
sscodeai/ai-project-skill-maker - License: Apache-2.0
- Runtime: Node.js scripts plus platform-neutral Markdown templates
- Install target:
~/.codex/skills/ai-project-skill-maker
It is designed to be platform-neutral. The default template works as an AI project maintainer skill, and the reference adapters describe how to emit Codex skills, AGENTS.md, CLAUDE.md, Cursor rules, and .github/copilot-instructions.md.
Two modes
| Mode | For | Inputs |
|---|---|---|
| Genesis Mode | A new project with no meaningful repository evidence yet | language and locale preferences, product goals and non-goals, target users, technical constraints, documentation voice, quality bar, release posture, first milestones |
| Repo Mode | An existing repository | README* / docs / contribution / changelog / license files, package.json and lockfiles, build and test configs, CI workflows, scripts and tests and entry points, generated-file signals, commit and style signals |
Genesis output is a day-one maintainer skill that records declared intent and recommended standards — without pretending they are observed repository facts. Repo output is a long-term maintainer skill grounded in observed project facts.
Language and locale setup
Genesis mode starts by asking what language the author thinks and reviews best in, then asks whether the rest of the interview and the generated references should use that language. For example, Japanese projects can choose です・ます, だ・である, or a mixed technical documentation style. When there is no preference, the default is English.
Evidence model
Generated project references separate claims into:
| Label | Meaning |
|---|---|
observed_fact |
Evidence from repository files, commands, CI, tests, metadata, or history (cites the file path where possible) |
declared_intent |
Goals, constraints, and preferences explicitly provided by the maintainer |
recommended_standard |
Standards selected because they fit the project type |
inferred_assumption |
Careful assumptions made when information is missing |
Generated output
project-maintainer/
SKILL.md
agents/openai.yaml
references/
project-intent.md project-map.md architecture.md
coding-standards.md content-style.md workflows.md
verification.md release.md generated-files.md
scripts/health-check.* optionalGenerated Markdown files contain a hand-written block that is preserved on refresh:
<!-- BEGIN USER RULES -->
<!-- Add durable project-specific rules here. This block is preserved on refresh. -->
<!-- END USER RULES -->Scripts
node scripts/collect-repo-signals.mjs /path/to/repo > repo-signals.json
node scripts/draft-project-config.mjs --repo /path/to/repo > config.json
node scripts/validate-config.mjs --input config.json --mode repo --strict
node scripts/render-project-skill.mjs --input config.json --output ./my-project-maintainer --strict
node scripts/validate-project-skill.mjs ./my-project-maintainerRender platform-specific instruction files:
node scripts/render-adapter.mjs --input config.json --adapter agents --output .
node scripts/render-adapter.mjs --input config.json --adapter claude --output .
node scripts/render-adapter.mjs --input config.json --adapter cursor --output .
node scripts/render-adapter.mjs --input config.json --adapter copilot --output .--init-config genesis|repo prints a starter config and --print-schema prints the schema. Forward-test prompts for fresh-session evaluation live in references/evals/forward-tests.md.
Current capabilities
- genesis mode for intent-first new projects, and repo mode driven by observed repository signals
- language and locale setup for author interviews and generated references
- config schema, example configs, and strict config validation (evidence labels and repo fact citations)
- rendering and validating project maintainer skill folders
- refresh workflows that preserve user-authored rule blocks
- local Codex personal skill installation and installed-content self-checks
- adapter guidance for Codex skills,
AGENTS.md,CLAUDE.md, Cursor rules, and.github/copilot-instructions.md - Astro and documentation sites, TypeScript and Node OSS projects, documentation-heavy projects
Next
- Read the README and
references/in the repository - See general skill generation in ai-skill-maker
- Browse other SSCodeAI projects in Projects