A table maps bugs, parsers, adapters, performance, documentation, and pull requests to focused documents. It gives an entry path into the material without loading every guide for every change.
AGENTS.mdL62–69
Source excerpt starting at line 62.
62| Task | Guide |63| ----------------------- | --------------------------------------------------------------------------------- |64| Fix a bug | See [Testing Patterns](.agents/docs/testing.md) → Regression |65| Add a new parser | See [Parser Implementation](.agents/docs/parser-implementation.md) |66| Add a framework adapter | See [Adapter Development](.agents/docs/adapter-development.md) |67| Improve performance | See [API Design](.agents/docs/api-design.md) → Performance & Reliability |68| Update documentation | See [Release & Git Workflow](.agents/docs/git-workflow.md) → Documentation |69| Prepare a pull request | See [Release & Git Workflow](.agents/docs/git-workflow.md) → PR Quality Checklist |
The skill workflow asks the agent to inspect descriptions first and read complete instructions only for relevant tasks. An explicit exclusion keeps unrelated domain guidance out of the working context.
AGENTS.mdL60–67
Source excerpt starting at line 60.
60Domain-specific context and conventions live in `.agents/skills/`. Before61starting any task:62631. List `.agents/skills/*/SKILL.md`642. Read only each skill's frontmatter description to identify relevant skills653. Read only the full `SKILL.md` files relevant to your task6667Do not load skills that are not relevant to the task.
Nine skills each have a clear trigger, including user-facing text, docs work, SQL built against user databases, and form code. Those skills, not the root file, own the detailed conventions.
AGENTS.mdL63
Source excerpt starting at line 63.
63The skills in `.agents/skills/` are the source of truth for conventions — load the relevant ones before working, don't guess:
nuqs 
Pierre 
Supabase 
