projectbluefin/documentation/docs

Skill router

Task → skill. Load only what the task needs.

Read this first, whatever the task: published prose is human-authored. Agents format, structure, and embed it — they never invent narrative, history, motivation, or first-person statements under a maintainer's byline. See `AGENTS.md` → _Never write in a maintainer's voice_. That rule outranks any writing guidance in a general-purpose design or content skill you have loaded.

Repository rules, build commands, git workflow, data-pipeline contracts and presentation rules all live in AGENTS.md at the repository root. It is authoritative; this page only routes.

Adding a skill

A skill earns a file when it is non-obvious, repeatable, and would otherwise be rediscovered by the next agent. One page per skill, in docs/skills/, added to the table above in the same pull request.

Every skill file carries front matter with a name and a description containing "Use when" trigger phrases, then these sections:

## When to Use              triggering conditions
## When NOT to Use          exclusions, with links to the skill that does apply
## Core Process             the numbered workflow
## Common Rationalizations  the excuse, and why it is wrong
## Red Flags                concrete signs the skill is being violated
## Verification             checkable exit criteria
## Sources                  files, and Context7 library IDs for verified content

Technical claims about a library, framework, or CLI are verified through Context7 before they land, and the library ID is recorded under ## Sources.

Note that docs/ is mounted at routeBasePath: "/", so everything here publishes to [docs.projectbluefin.io](https://docs.projectbluefin.io/). Write skills for a public audience. Internal design records belong in adr/ at the repository root, which does not publish.