ai-agent-skills

Produce the source docs an app feeds into the LLM Wiki (processor) — a compliant README for any kind (API / cronjob / worker / knowledge), plus a fresh openapi.json for OpenAPI-capable apps. Self-contained, bundled stdlib tools.

Convert an operations SOP (DBA runbook, infra procedure, deployment checklist) into a spec a human can approve and an AI agent can implement a three-layer FastAPI service from.

Add or modify an installable Claude Code skill in this repo — produce a spec-compliant SKILL.md, bundle its scripts/references, and register it in the marketplace. Use when authoring a new skill or editing an existing one.

Develop three-layer FastAPI services on the company API template — start a new service by cloning the template (architecture, devcontainer, lint/CI and built-in utilities like DB clients and mailers already in place), then add endpoints following the layered architecture, reusing built-in tools instead of reinventing them. Template specifics live in a regenerable facts layer with a rescan procedure.

Bundle — installs ALL of this repo's own skills at once (the full list is this entry's skills array; new skills added to it arrive automatically via marketplace auto-update).

External — brainstorming, subagent-driven dev with code review, systematic debugging, red/green TDD, and skill authoring. Mirrored from obra/superpowers into the internal GitLab; swap the url to your GitLab mirror. No sha = follows the mirror's default branch, so `/plugin marketplace update` pulls whatever the mirror last synced.

External — behavioral guidelines to reduce common LLM coding mistakes (Think Before Coding, Simplicity, …). Mirrored from forrestchang/andrej-karpathy-skills into the internal GitLab; swap the url to your GitLab mirror. No sha = follows the mirror's default branch.