[LAYER1-SIM] SKILL.md regression detected by Claude simulation

[github-actions]

The Layer 1 Claude simulation harness at tests/harness/layer1-claude-sim.test.ts detected a SKILL.md regression: Claude's response did NOT mention the expected bun run shape one of the three skills prescribes.

Possible causes:

• SKILL.md prose became ambiguous (a refactor stripped the explicit Bash invocation example).
• Claude model behavior drift on the pinned ANTHROPIC_HARNESS_MODEL.
• Skill body referencing a script path that no longer matches the runner.

Next steps: rerun the workflow with workflow_dispatch to confirm reproducibility; review the failing SKILL.md against the captured response in the workflow log; ship a SKILL.md prose fix.

[tgorka]

Triage — false alarm, fixed in PR #73

This issue fired because the workflow's ANTHROPIC_API_KEY env var was set to an empty string (GitHub Actions behavior when a referenced secret isn't configured). My harness's HAS_API_KEY = process.env.ANTHROPIC_API_KEY !== undefined check evaluated TRUE → tried to call Claude with an empty key → HTTP 401 → test failed.

Fix in PR #73 (tests/harness/layer1-claude-sim.test.ts + .github/workflows/layer1-sim.yml): the harness now also checks length > 0, and the issue-opener step is gated on env.ANTHROPIC_API_KEY != '' as defense-in-depth.

This is a Stepper CI bug, not a real SKILL.md regression. Whose problem? Mine (the workflow). The bmad-plugin had nothing to do with it.

Closing once PR #73 merges.