Skip to content

v3.22.0 - Skill-Support Delivery + Verify-at-Point-of-Use + Hygiene

Latest
Latest

Choose a tag to compare

View all tags
@gmelli gmelli released this 13 Jun 16:32
· 2 commits to main since this release
v3.22.0
ff1c72d

Theme: Skill-Support Delivery + Verify-at-Point-of-Use + Hygiene

For: downstream agent maintainers deciding whether to adopt v3.22.0.

No breaking changes. Minor, additive governed tooling. Existing instances upgrade by version-bump only. Two principles ship as runnable gates. First, propagation carries a paired point-of-use verify — the freshness and coherence checks are that verify half. Second, distribution stays surgical — the skill deployer works one skill at a time, not as a blanket overwrite.

What's New

  • Surgical skill deployment lands as a governed script. scripts/deploy_skill.py deploys one skill at a time with its companion artifacts, defaults to a dry run, logs every change for audit, and runs an independent post-deploy verification.
    Traceability: deploy_skill.py

  • Citation freshness becomes a runnable gate. scripts/check_claim_freshness.py parses governed artifacts for #NNNN STATE issue-citations; --online re-derives each via gh; --strict exits nonzero for CI and release gating.
    Traceability: check_claim_freshness.py

  • Skill-frontmatter coherence becomes a runnable gate. scripts/check_skill_coherence.py detects declared-but-absent Governing-Spec or SOP references, plus stale "future" framing of artifacts that already shipped.
    Traceability: check_skill_coherence.py

  • Ontology-spec binding gains its validator half. scripts/validate_spec_binding.py validates ONTOLOGY-SPEC-BINDING, with scripts/ground_artifact.py as its loader dependency.
    Traceability: validate_spec_binding.py, ground_artifact.py

  • Test coverage grows by 20-plus V-tests across the new scripts, all passing in canonical.

Changed

  • /aget-propose-actions advances to v1.8.0. It adds Step 3.5 Self-Critique (a 10-point checklist), REQ-PA-018/019/020 (named-person leverage and a non-artifact action at multi-day budgets), and a Type column that makes action-type bias visible. Propagated to all 13 templates, also clearing the accumulated v1.7.x lag.
  • scripts/validate_ontology_reciprocity.py scope fix. Dangling now means whole-vocabulary; cross-cluster is reclassified as non-defect and excluded from TOTAL DEFECTS.

Sleeping-CAPs Disclosure

Zero new sleeping CAPs at the V-test layer in v3.22.0. No prior-cycle sleeping CAPs were closed or grace-extended by this release.

Compatibility

No breaking changes in v3.22.0. Existing instances upgrade by version-bump only.

See CHANGELOG.md for the full Added and Changed sections.

What This Release Doesn't Change

  • The five pre-existing test failures (contract-test reference-rate, config tier-detection, skill-count map) are carried debt, disclosed not hidden. They were not introduced by this release.
  • Skill deployment shipped as a script; the broad deploy across downstream instances is tracked separately and is not completed by this tag.

Traceability: CAP-REL-006-02-NN release-body template; H-AMB-322 ambition experiment.

Assets 2
Loading