Skip to content

Roadmap

The work is broken into two tracks: skills (content the agent reaches for) and governance (the lifecycle that keeps that content current).

Skills track

Shipped

  • 15 developer skills covering the common Cardano workflows: scaffolding, writing validators, security review, optimisation, building transactions, designing tokens, debugging, querying chain data, devnet setup, wallet integration, governance, and the conceptual primers (eUTxO, CIPs, tooling).
  • The cardano-context skill, which writes a durable per-project directive into CLAUDE.md so agents reliably consult bundled context.
  • 48 documentation sources mirrored locally — SDKs, languages, infra, CIPs, ledger specs — under docs/sources/.
  • A SessionStart hook (hooks/check-docs.sh) that reports doc freshness and surfaces the per-project directive nudge.

Planned

  • Auto-consultation hook. A UserPromptSubmit hook that scans the user’s prompt for Cardano-specific keywords (aiken, plutus, cip-XXXX, ogmios, drep, …) and injects an “additional context” reminder to consult bundled docs/skills before training data or the web.
  • Usage telemetry. A PostToolUse hook logging which docs and skills were consulted per session, to a local file. Used to tune the keyword set and skill triggers based on real prompts.
  • New skills as the ecosystem evolves. New CIPs, new SDK paradigms, new validator patterns. Proposals via issue, ship via PR.

Governance track

Shipped

  • Weekly upstream refresh. Every Monday 06:00 UTC, all sources are re-fetched and a PR is opened with the diff. Human review + merge before content lands on main.
  • Manifest self-healing. The fetch script derives .manifest.yaml from disk state after every fetch (partial or full), so the manifest can’t drift from reality.
  • Schema validation. CI runs scripts/validate.py on every PR touching skills/** or registry/**.
  • Auto-derived counts. scripts/update-doc-counts.sh rewrites sentinels in CLAUDE.md and README.md from disk. CI runs --check to fail PRs on drift.
  • Source-vetting bar. Explicit policy in CONTRIBUTING.md: last commit age, release/activity signal, archival status, fork canonicality.

Planned

  • PR-time source-build check. When registry/sources.yaml changes, CI fetches the touched source(s) and verifies the clone + glob patterns produce files. Catches dead repos and bad globs before they land.
  • AI-powered governance review. On PRs touching registry/, skills/, hooks/, or scripts/, an AI reviewer reads the diff and the rules from CONTRIBUTING.md and posts an advisory verdict comment (not blocking — humans still merge).
  • Cross-tool compatibility surface. Codex and other agent harnesses consume the same skill files via .agents/skills symlinks. As the Agent Skills standard evolves, we follow it.

The principle across both tracks: ship small, observe, iterate.