Public Skills

This repository contains reusable skills for AI agents. Each skill is in skills/<skill-name>/.

Install a skill

Replace <skill-name> with a name from the catalog:

npx skills add btfranklin/skills --skill <skill-name> -g

Omit -g to install the skill only in the current project.

Skill catalog

The description field in each SKILL.md is the authoritative source for the skill trigger, scope, exclusions, and output.

Skill
deep-code-elegance-and-beauty-review
deep-codebase-review
design-html-first-web-uis
design-ui-style-guide
django-6-upgrade-guide
django-pytest-performance-suite
integrate-daisyui-into-django
llms-txt-authoring
make-repo-agent-legible
openai-django-webhooks
publish-npm-package
publish-python-package-pypi
review-agent-production-readiness
skill-icon-workflow
structured-llm-output

Repository workflow

Use these files for each skill:

• Each skill must contain SKILL.md. Its frontmatter name must match the skill directory. Its description is the authoritative trigger and scope.
• Each skill in this repository must contain agents/openai.yaml. This file contains the display name, short description, default prompt, and icon paths.
• assets/ must contain the SVG and PNG files named in agents/openai.yaml.
• Each skill must contain LICENSE.txt. This file permits separate distribution of the skill directory.
• references/, examples/, and templates/ are optional. Add them only when the skill needs progressive disclosure or reusable material.
• evals/evals.json is optional for a simple skill. Add it when trigger boundaries, approval gates, fallbacks, or failure behavior need regression coverage.

Keep detailed behavior in one authoritative file. Link to that file instead of copying its content into the README or another skill file.

Writing standard

Use ASD-STE100 Simplified Technical English for all technical prose. Follow the rules in AGENTS.md. Review the language during normal editing and review. The repository validator does not enforce this writing standard.

Before you publish a change, run:

python3 scripts/validate_repository.py

The validator checks repository structure, frontmatter, metadata, icon paths, local Markdown links, eval JSON, and the README catalog. It validates the eval file structure, but it does not execute behavioral evals. Run behavioral evals in the agent harness that will use the skill.

Run the deterministic example checks with PDM:

pdm sync -G test
pdm run test

These checks execute the published invoice schema and arithmetic code. They also execute the webhook endpoint function with a SQLite receipt store and a controlled queue. They cover queue failure after commit, repeated delivery, uncertain queue acceptance, missing invoice values, currency conflicts, and arithmetic errors. Specific instruction checks protect the database-engine rule and optional-skill fallback. These are contract checks, not automated writing-style checks.

CI runs both the repository validator and these checks. The checks do not call a model or an external service. They do not replace Django integration tests, provider tests, or agent-level evaluations from evals/evals.json. Run those evaluations in the target agent harness and record the inputs, tool calls, results, and unmet expectations. Do not report the deterministic checks as proof that an agent followed the instructions.