Roadmap

Status: protocol Draft 0.5.0, evolving in the open via RFCs; the @skillerr/* reference packages implement it as it stands today. The goal is an open, interoperable contract — multiple conforming hosts and runtimes, not a single CLI.

Public, contribution-oriented items only. Timing is not guaranteed. Site: skillerr.com.

Available in the reference today

• .skill container + digests
• Mint + creation attestation (reference implementation)
• Free local transparency-log registry
• Reference runtime + CLI (skillerr / skill)
• Conformance tests, docs, governance, CI
• Local / offline agent provenance (Ollama, LM Studio, llama.cpp, custom)
• Transferable SkillContract, assessment APIs, and JSON Schema
• Structured contract-to-manifest / workflow compilation
• Production-grade signing — real Ed25519 issuer keys + a local trust store (skill keygen, skill mint --signer-key, skill verify-trust --trust-store), not just the bundled development-only key
• Adversarial package corpus (zip bombs, path tricks, hash mismatch, and more) — runs on every push in CI, not just described here
• Public RFC folder (docs/rfcs/) — six RFCs; asymmetric signing has since shipped as real code, the rest remain spec-only proposals open for discussion
• Optional public transparency-log anchoring (skill mint --transparency), built on the official sigstore/Rekor client libraries — see Trust and security
• Fulcio keyless mint (skill mint --keyless) — an OIDC-bound identity anchor, zero setup inside GitHub Actions
• Independent Rekor verification link on every anchored trust check, in both the CLI and the verify page — a link to sigstore's own public log, not just this tool's word
• Per-claim assurance model (skill inspect --trust --claims) — every claim split into two structurally separate lists, verified vs. self-reported, so nothing can misrepresent one as the other
• License/terms manifest slot (SPDX identifier + terms URL)
• Public verify utility on this site — upload a .skill file, get the same TrustView skill inspect --trust reports, without installing anything

Next (good contribution targets)

• Independent implementation validating the published authoring schema
• Interactive/browser-login OIDC provider for skill mint --keyless run locally, outside CI (the CI-ambient path already ships)
• HTTP transparency-log server (same log format as the local registry)
• Stronger verify assertion language + fixtures
• Host adapters: local OpenAI-compatible, Cursor, Claude Code, Codex
• Second language runtime (Go or Rust) for Stable eligibility
• Official SKILL.md round-trip adapter tests

Later

• Multi-issuer trust roots / key transparency
• Optional ledger anchors as one permanence kind (never required)
• Mark Candidate, then Stable, after two independent runtimes pass the same corpus

Stability path

Draft → Candidate → Stable requires independent conforming runtimes and a shared adversarial / conformance corpus — not a single vendor ship date.