SkillsBar

OpenAI Hackathon · Interactive prototype

SKILLSBAR · BY brAInwav

When evidence goes stale.

A changed Skill candidate should not inherit proof that no longer belongs to it. SkillsBar shows what a maintainer can safely conclude, and the exact evidence to inspect next.

CURRENT LOCAL CANDIDATE

Needs candidate identity

1 receipt current · downstream proof held

11/100pipeline readiness
01

NEXT REQUIRED

Candidate identity

Build the canonical non-mutating package receipt.
  1. 01Candidate identityCanonical package digestRequired
  2. 02Mechanical validationPackage verify and strict auditHeld
  3. 03Security reviewCandidate-bound risk modesHeld
  4. 04Eval preparationScenarios and scorer calibrationHeld
  5. 05Local evaluationGoverned local eval receiptHeld
  6. 06Cloud evaluationRemote eval receiptHeld
  7. 07Tessl stagingCandidate-bound staging receiptHeld
  8. 08Runtime proofInstalled digest and observed behaviorHeld
  9. 09Registry publicationPublished version and visibilityHeld

WHY CODEX MATTERS

From repository evidence to a release decision.

  1. 01Collect

    Codex runs the Skills SDK’s read-only checks and preserves structured receipts.

  2. 02Bind

    Every downstream result must match one canonical package digest and scenario set.

  3. 03Challenge

    Security, eval, and runtime evidence remain independent instead of becoming one vague green tick.

  4. 04Act

    SkillsBar exposes the next exact command, so an operator can fix the weakest gate first.

HUMAN-OWNED DELIVERY

Built with Codex. Directed and verified by Jamie.

Jamie
Framed the stale-evidence problem, set the product direction, made taste decisions, and owns every public claim.
Codex
Implemented bounded changes, ran repository checks, preserved structured evidence, and supported inspection.

THE IDEA

Release confidence should be inspectable, not inferred.

Safe mock data · No credentials · No production endpoints