docs(audit): vibe-code-audit v2.1 + codebase maps (2026-04-17)

Artifacts under docs/audit/2026-04-17/:
  AUDIT.md          - full 13-section due-diligence audit
  FIX_PROMPTS.md    - 7 paste-ready AI fix prompts for Critical/High
  FEATURE_IDEAS.md  - ranked feature proposals
  codebase/         - gsd-codebase-mapper outputs (7 files):
    STACK.md, INTEGRATIONS.md, ARCHITECTURE.md, STRUCTURE.md,
    CONVENTIONS.md, TESTING.md, CONCERNS.md

Vibe Debt Score: 42/100 - FIX BEFORE LAUNCH.
Top risks: crash loop (systemd PIDFile vs puma -d mismatch),
plaintext secrets in systemd unit, PipelineController admin fallback,
HOOKS_TOKEN unset in production, webhook idempotency missing.

No hard-stops found. Auth layer, CI pipeline, rate limiting are
stronger than the vibe-coded framing suggested.

v2026.3.2 — New models, pipeline cleanup, auto-runner fixes

Highlights:
- Add gpt5.4mini and gpt5.4nano to Task::MODELS
- Update OPENCLAW_MODEL_ALIASES (codex->gpt-5.4, new mini/nano)
- Fix recurring_templates scope to exclude needs_decision status
- Fix agent_complete hook description rebuilding
- Remove 141 lines of dead pipeline code (Task, AutoRunner, WebhookService)
- Fix auto-fill origin_chat_id from user.telegram_chat_id
- Fix post-audit #858: task.title NoMethodError, nil guards
- PinchChat-inspired UI: tool badges, thinking blocks, token bar

Compatible with OpenClaw v2026.3.22

feat: health gate before autopull + delivery backoff retry (closes #22)

docs: update CHANGELOG to 2026.3.0 and README feature list

Factory v2 + ZeroBitch Fleet + Kanban Performance

🏭 Factory v2 — Continuous improvement engine with 10 built-in agents
⚔️ ZeroBitch — Fleet management dashboard for ZeroClaw instances
⚡ Kanban Performance — N+1 fixes, fragment caching, Turbo frames
🔧 Cron Builder — Edit/delete/delivery target improvements
🧪 30+ new tests across models, controllers, jobs, services
🐛 15+ bug fixes (zerobitch, mobile, factory, memory search)