Create a typed agent skill registry with usage counters #90

Outcome

Create a registry for digest-pinned, typed agent skills that records verified capabilities and usage without promising royalties before settlement works.

Current behavior

Skills and plugins carry the proposed network moat, but no current surface lists a skill's input, output, compatibility, evidence, version, or use. Issue #35 provides a governed extension pattern, and project 8 is building the payment prerequisite.

Source: docs/episode-triage.md, episodes 206, 238, 269, and the recruitment station.

Contract

• Register a skill name, immutable version digest, typed input and output, author, source and license, supported executors, authority requirements, cost policy, and verification receipts.
• Keep mutable discovery metadata separate from immutable skill versions.
• Count admitted invocations, successful accepted outcomes, failures, and paid uses separately.
• Bind usage to work and outcome receipts instead of trusting client counters.
• Expose search and read APIs before enabling third-party execution.
• Require review and sandbox policy for executable artifacts.
• Preserve export and removal while retaining bounded historical receipts.
• Do not add FROST copy protection or hosted custody.

Acceptance criteria

• A client can discover a skill and determine whether its types, authority, executor, and evidence fit a workflow.
• Two different artifacts cannot publish under the same version digest.
• Usage counters reconcile with durable work receipts and distinguish attempted from accepted use.
• Removing a listing blocks new routing without erasing prior evidence.
• No royalty or payout state appears until the settlement dependency ships.

Verification

Test digest identity, types, search, authorization, usage attribution, removal, private skills, malicious metadata, and executable-artifact admission.

Dependencies

Depends on #35, project 3 accepted-outcome receipts, and project 8 bounty settlement.