The cross-model consistency matrix (models_to_test) in 54 skill evals + both
eval templates pinned retired/malformed model IDs — `claude-3.5-sonnet`
(malformed: dashes, not dots; resolves to Sonnet 3.5, retired Oct 2025) and
`claude-3-haiku` (retired Apr 19 2026). Both 404, so those evals could not run
a meaningful cross-model check.
Refresh to current pinned IDs, preserving the original "primary + fast floor"
design:
- claude-sonnet-4-6 # primary (high accuracy)
- claude-haiku-4-5 # fast floor (minimum quality)
High-stakes skills additionally get the capability ceiling claude-opus-4-8
(security-testing, security-visual-testing, n8n-security-testing,
pentest-validation, chaos-engineering-resilience, qe-chaos-resilience,
qe-defect-intelligence) and the security eval template. Opus is scoped to these
7 + the security template rather than all 54 to avoid a fleet-wide eval-cost
blowup.
Also fixes the two `--model claude-3.5-sonnet/claude-3-haiku` examples in
run-skill-eval.ts and a nested cross-model test-case input in
qe-learning-optimization.yaml.
Verified: all eval YAML parses; zero retired/deprecated Claude IDs remain across
skills/evals + templates + runner. (src/domains/constants.ts MODEL_TIERS was
already current — tier 1 claude-haiku-4-5 — so it is intentionally unchanged;
the legacy claude-3-haiku-20240307 entries in cost-tracker/bedrock are retained
historical pricing/availability data.)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Implement three P1 items from skill-evaluation-research.md that deepen
eval grading in the runner's validateOutput method:
- P1-6: Negative control test cases (inverted pass logic for "skill
should decline" tests)
- P1-5: Finding count enforcement (JSON parse + regex fallback),
schema_path validation, and weighted grading rubric scoring
(completeness/accuracy/actionability)
- P1-4: Adaptive rubric that extracts keywords from prompts (quoted
strings, format words, named standards) as dynamic must_contain checks
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(v3.4.1): bundle missing dependencies and test timeouts
- Fixed issue #219: MCP bundle failing due to missing packages
- Changed build scripts from --packages=external to selective externalization
- Pure JS deps (fast-json-patch, jose, uuid) now bundled inline
- Native modules properly externalized (better-sqlite3, hnswlib-node, etc.)
- CommonJS modules with dynamic requires externalized (typescript, fast-glob, yaml, commander, cli-progress, ora)
- Bundle size reduced from ~15MB to ~5MB
- Added timeouts to handleQualityAssess tests to prevent CI failures
- Updated version to 3.4.1
Closes#219
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat(v3.4.2): add skill validation system with trust tiers (ADR-056)
Implement 4-layer skill validation system for deterministic outputs:
- Tier 3 (Verified): 46 skills with full eval test suites
- Tier 2 (Validated): 7 skills with executable validators
- Tier 1 (Structured): 5 skills with JSON output schemas
- Tier 0 (Advisory): 5 skills with SKILL.md guidance only
Key additions:
- JSON Schema validation templates for skill outputs
- Executable validator scripts for correctness verification
- Evaluation test framework with multi-model support
- CLI commands: aqe skill report/summary/compare, aqe eval run/status
- ReasoningBank integration for validation pattern learning
- GitHub Actions workflow for CI skill validation
- Trust tier frontmatter in all 63 skill SKILL.md files
- Comprehensive documentation and user guides
Infrastructure:
- .claude/skills/.validation/ - Templates and frameworks
- .claude/skills/*/schemas/ - Output schemas per skill
- .claude/skills/*/scripts/ - Validator scripts per skill
- .claude/skills/*/evals/ - Test cases per skill
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(ci): trust tier-based skill validation
- Tier 3 (Verified): Blocks PR on validation failure
- Tier 2 (Validated): Warns only, doesn't block
- Tier 1 (Structured): Warns only, doesn't block
- Tier 0 (Advisory): Skipped entirely
Reads trust_tier from SKILL.md frontmatter to determine
validation requirements. Only 46 Tier 3 skills block merge.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(validators): remove local keyword from self-test blocks
Shell script fix: 'local' can only be used inside functions, not in
if/else blocks. Both accessibility-testing and compatibility-testing
validators had 'local' declarations in their self-test sections that
caused errors when run outside of a function context.
- Remove local test_file from accessibility-testing self-test
- Remove local test_file, browser_count, tier1_pass from compatibility-testing self-test
- Sync fixes to v3/assets/skills/
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(validators): add --self-test support to n8n skill validators
The 5 n8n skill validators were failing CI because they didn't handle
the --self-test flag properly. Added self-test handling that:
- Checks for required tools (jq)
- Verifies schema file exists and is valid JSON
- Returns exit 0 on success
Fixed validators:
- n8n-expression-testing
- n8n-integration-testing-patterns
- n8n-security-testing
- n8n-trigger-testing-strategies
- n8n-workflow-testing-fundamentals
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Dragan Spiridonov