Cardog Agent Skills

The cardog skill teaches any skills-compatible agent harness — Claude Code, Cursor, Codex, Copilot, Goose, and the rest — how to use the Cardog platform correctly: resolve free text once, hold permanent refs, read the response envelope, recover from errors in one turn.

Install

From the platform (discovery via .well-known):

npx skills add https://cardog.app

From this repo:

npx skills add cardog-ai/agent-skills

Both resolve to the same artifact.

One truth, three surfaces

The skill is generated inside the cardog.app application from @cardog/contracts (the OpenAPI source) and @cardog/entities (the ref grammar). It is served at cardog.app/skill.md, indexed with a sha256 digest at /.well-known/skills/index.json, and mirrored here.

This repo cannot drift: CI verifies the repo copy byte-matches the digest-verified live artifact on every push and PR, and a daily job opens a sync PR when the platform ships a new skill. See scripts/sync.sh.

What the skill covers

• Resolve-first — free text enters the API at exactly one endpoint; everything else takes a ref or a VIN.
• The ref grammar — {domain}:{key}, permanent join keys, validated offline by @cardog/entities.
• Job routing — identity, inventory, market, recalls, resolution: which call answers which job.
• Envelope self-correction — read code, follow hint, apply suggestions.
• One-fetch docs — curl cardog.app/docs.md before asking anything.
• Auth rails — keys, the X-Credits-* headers, what a 402 means.

Full platform docs: cardog.app/docs (agents: cardog.app/docs.md).

License

Apache-2.0