Session Triage: provision-agentic-config Benchmark Failure

Generated 2026-05-22 for benchmark/triage-provision-agentic-config-2026-05-22.md

Verdict

Partially verified. The benchmark failure is real, but the retained artifacts show a benchmark false negative rather than a verified skill-output defect.

Evidence

Both agents exited successfully and wrote AGENTS.md.
Retained artifacts include ## Workflow Orchestration, ### 4. Verification Before Done, ### 7. Monorepo Parallel-Work Safety, shipping rules, no-GitHub-Actions guidance, and Recommended next command: $exec.
The fixture expected literal shorthand phrases orchestration rules and monorepo safety, which are not the canonical headings required by the skill contract.

Root Cause

The benchmark setup reuses shorthand prompt terms as exact artifact assertions and quality facts. The generated policy block uses canonical section names, so correct outputs fail phrase-level checks.

A secondary quality false positive flags package-lock.json, even though the required monorepo safety block legitimately names shared lockfiles.

Recommended Fix

Route to $targeted-skill-builder provision-agentic-config benchmark fixture false negative.

Update tests/layer4/setups/tier23-global-workflows.setup.ts to assert canonical headings and evidence patterns.
Allow package-lock.json for this fixture.
Add focused layer1 coverage proving canonical policy sections pass and shorthand-only echoes fail.