A personal collection of Agent Skills — portable, version-controlled instructions that extend skills-compatible agents (Claude Code, Cursor, Codex, and others).
Install every skill into your agent with the skills CLI:
npx skills add fm1randa/skillsOr a single skill by its path:
npx skills add https://github.com/fm1randa/skills/tree/main/skills/code-review-pyramidThe CLI installs a copy into your agent's skills directory rather than symlinking, so re-run the command (or npx skills update) to pull in updates.
| Skill | What it does |
|---|---|
| code-review-pyramid | Code review organized by Gunnar Morling's Code Review Pyramid — parallel per-layer agent fan-out, confidence-scored findings, severity capped by pyramid layer. |
| review-loop | Runs rounds of local pre-merge review, one fresh and disposable reviewer sub-agent per round, until a round is clean or two rounds report nothing blocking. |
| recon | Gathers context on an Agidesk task by fanning out one researcher agent per source (Agidesk, Slack, Google Drive, and Datadog for bugs) into a single context brief. |
| open-pr | Opens a pull request with a short, high-level description following a fixed convention — confirms the base branch and links the related task as Fixes/Closes/Addresses. |
| feedback | Drafts a short, low-jargon findings message in Brazilian Portuguese as raw HTML, following a fixed template with hyperlinked PRs. |
| use-flpm-ui | Teaches agents to build React UI with FLPM UI, the personal shadcn registry on Base UI: project setup, namespaced installs, Base UI API differences, visual rules and button hierarchy. |
| output-language | Locks the reply language of a session instead of inferring it from the conversation. Named profiles and a default live in ~/.config/output-language, so every session follows the default with no skill run; /output-language <profile id | free text> pins one session, default makes it inherit again, off disables it. An always-on hook re-injects the choice every turn, so it survives loading other skills. Needs a one-time hook setup. |
This repo is the source of truth — author and commit skills here. The skills CLI installs a copy into your agent's skills directory (for Claude Code, ~/.claude/skills); it does not symlink to this working tree, so re-sync after editing:
git clone https://github.com/fm1randa/skills ~/Repositories/skills
cd ~/Repositories/skills
# Install / refresh from your local working tree (Claude Code only):
npx skills add ~/Repositories/skills --agent claude-code -g -yBecause the CLI copies rather than symlinking to the working tree, re-run that command (or npx skills update) after each edit to pick up changes. See CLAUDE.md for authoring conventions and GLOSSARY.md for the vocabulary.
The code-review-pyramid skill bundles "The Code Review Pyramid" diagram by Gunnar Morling, © @gunnarmorling, licensed CC BY-SA 4.0 and included unchanged. See the attribution note.
MIT © Filipe Miranda. Bundled third-party assets retain their own licenses (see Credits).