Open-source AI tooling
Agent Skills
Reusable, evidence-backed agent skills for practical software development.
Problem and purpose
Agent instructions are easy to write but harder to make dependable. A useful skill needs a clear trigger, a bounded workflow, explicit safety rules, and an output another person can verify. Agent Skills provides a public home for developing those pieces together, starting with an audit for repositories being considered for public release.
Anatomy of a skill package
SKILL.md
Defines when the skill applies, the workflow to follow, safety boundaries, and evidence standards.
references/
Keeps the detailed audit rubric and report template available without overloading the core instructions.
agents/
Supplies agent-facing metadata so supported tools can present the skill consistently.
github-public-readiness execution flow
- Establish scope: inspect the repository, history, documentation, dependencies, automation, and available release artifacts.
- Find disclosure risks: use targeted checks for credentials, private material, internal references, production data, and problematic history.
- Check usability and hygiene: assess whether the project purpose, setup, license, tests, and maintenance state are understandable.
- Judge portfolio value separately: evaluate what the repository demonstrates without confusing showcase value with publication safety.
- Report evidence: give a readiness classification, concrete findings, uncertainty, and the smallest ordered release checklist.
Safety and evidence design
The skill is advisory by default. It does not change visibility, publish code, rewrite history, delete files, or rotate credentials unless that additional action is explicitly authorized. Possible secrets are treated as leads to verify, and any real credential is redacted from the report.
Findings are expected to point to file paths, commits, command output, or reproducible checks. Missing access and checks that could not run are reported as limitations instead of being silently converted into confidence.
A reusable foundation
The repository structure separates compact agent instructions from deeper references and output formats. That gives future skills a repeatable package shape while leaving room for plugin manifests that bundle related skills. Standalone plugins or MCP products can still move into dedicated repositories when their release and maintenance needs become distinct.
Install or generate a one-off prompt
Add the collection to a supported agent environment:
npx skills add erik-fryscok/skillsOr generate a one-off invocation prompt through the skills CLI:
npx skills use erik-fryscok/skills@github-public-readinessExample prompts
Use $github-public-readiness to audit this repository for safe public release and portfolio value.Audit this repository before I make it public. Prioritize disclosure risks and the smallest release checklist; do not modify files or visibility.Assess whether this repository is worth featuring in my portfolio, keeping public safety and showcase value as separate verdicts.Expected report output
The result includes a readiness classification, a separate portfolio judgment, evidence-linked findings, an ordered release checklist, and the verification performed. It also states confidence and important limitations so the recommendation can be reviewed rather than accepted on authority.
What this demonstrates
- Turning an ambiguous engineering review into a scoped, repeatable agent workflow.
- Designing safety boundaries that keep analysis separate from consequential repository changes.
- Making recommendations traceable to evidence, uncertainty, and reproducible verification.