Milestone #114 Bet B closed. Its 5 issues relabeled research. No drift signal has surfaced from shipped sites; Bet B revives only when a real drift case appears.
$ bash tests/security/test-npm-audit.sh ▶ root ✓ 0 findings ≥ moderate ▶ docs/site ✓ 0 findings ≥ moderate ▶ src/hooks ✓ 0 findings ≥ moderate Total: 3 | Passed: 3 | Failed: 0
Allowlist at .claude/audit-allowlist.json — entries carry until: dates so exceptions force themselves to be revisited.
src/agents/README.md (auto-generated) ### Decision tree — which agent for which task CI / workflow change? → ci-cd-engineer Python/backend service? → backend-system-architect React/Next.js component? → frontend-ui-developer Security audit / OWASP? → security-auditor Test generation? → test-generator Database schema? → database-engineer Infrastructure / IaC? → infrastructure-architect LLM / RAG / prompt work? → llm-integrator Explore codebase? → Explore (built-in) Unsure OR mixed concerns? → workflow-architect ### Agent registry (37 agents) | Agent | Category | Model | Effort | When to spawn | | ... | ... | ... | ... | ... |
Regenerated by node scripts/list-invocable-agents.mjs --write.
Adding src/agents/README.md as registry broke every counter that assumed "agent = *.md in src/agents/". Fixed 7 places so "agent" now means "file with agent frontmatter, not README / INDEX / CONTRIBUTING":
bin/count-components.sh, bin/validate-counts.sh, bin/validate-frontmatter.shscripts/stamp-counts.sh, scripts/build-plugins.shtests/ci/lint.sh, tests/agents/test-agent-frontmatter.sh, tests/performance/test-token-overhead.shtests/unit/test-count-components.sh, tests/indexes/test-index-generation.shWave 1 pre-push gate, commit-msg linter, token budget warn tier merged Wave 2 5 test debt PRs (formatStars, sync_versions, Playwright, schema validator, stub-fallback) merged Wave 3 agent registry, npm audit, Bet B → research ← THIS PR 7 dimensions × 10 = 10/10