chore: propagate Zoo Flow Cursor agent bundle from hub

Add AGENTS.md and sanitized .cursor rules/skills for agent workflows.

Author: jssturm

.cursor/rules/zoo-anti-loop-shield.mdc

@@ -0,0 +1,34 @@
+---
+description: Anti-Loop & Anti-Hallucination Shield (Zoo mandate)
+alwaysApply: true
+source: .roo/rules/11-anti-loop-hallucination-shield.md
+syncedAt: 2026-07-27T17:34:09.037Z
+---
+
+# Anti-Loop & Anti-Hallucination Shield
+
+## Anti-loop (max 2 attempts)
+
+Never execute the same terminal command or tool call more than twice if it errors or returns identical results.
+
+On second failure, output `<LOOP_BREAK>` with: what was attempted, errors from both attempts, why retries fail, alternate approach.
+
+| Failed | Pivot to |
+|--------|----------|
+| Test command | `grep`/`Read` to verify state |
+| `Read` on path | `Glob` on parent directory |
+| Complex `Grep` | Broader pattern or directory walk |
+| Build command | `ls` output dir + read error files |
+
+## Anti-hallucination
+
+Never reference a path unless observed via `Glob`/`Grep`/`Read` this session.
+
+Before `Write`/`StrReplace`/file-creating `Shell`:
+- **New file:** `Glob` parent directory
+- **Existing file:** `Read` target (even line 1)
+- **Command with paths:** verify each directory
+
+## Traceback logging
+
+Before tool calls on paths, note internally: target path and attempt number (1 or 2). On attempt 2, state what changed from attempt 1.

.cursor/rules/zoo-auto-invocation.mdc

@@ -0,0 +1,51 @@
+---
+description: Zoo auto-invocation policy — public repo safe (no proprietary Athena references)
+alwaysApply: true
+---
+
+# Auto-Invocation Policy
+
+**Everything in Zoo Flow is automatic.** The user does not type slash commands. Infer intent, load the matching skill, adopt the right role, and execute — in the same turn when safe.
+
+## Rules
+
+- All `.cursor/rules/zoo-*.mdc` mandates are always active
+- Never tell the user to run `/zoo-*` to proceed
+- Never wait for workflow approval when intent is clear — start executing
+- **Never expose proprietary Project Athena content** — see `zoo-no-proprietary-exposure.mdc`
+
+## Skills
+
+- All `.cursor/skills/zoo-*/` skills are **model-invoked**
+- When a task matches a skill's description, read and follow that skill's `SKILL.md` immediately
+- Do **not** load `zoo-athena-doctrine` in this public repo
+
+## Role adoption (automatic)
+
+| Task signal | Role | Skill(s) to load |
+|-------------|------|------------------|
+| Small known fix, CSS, typo | Code Tweaker | `zoo-tweak` |
+| Test-first, red-green | Code Tweaker | `zoo-tdd` |
+| Bug, broken, failing, debug | System Architect → Tweaker | `zoo-diagnosing-bugs` |
+| New feature, capability, PRD | System Architect | `zoo-grill-with-docs`, `zoo-to-spec`, `zoo-to-tickets` |
+| Refactor, architecture | System Architect | `zoo-improve-codebase-architecture` |
+| Unfamiliar code, explore | System Architect | `zoo-explore` |
+| Verify, tests pass, lint | Code Tweaker | `zoo-verify` |
+| Review diff/PR | System Architect | `zoo-code-review` |
+| Task completion | Code Tweaker | `zoo-completion-protocol` |
+| Commit, document | Code Tweaker | `zoo-commit-and-document` |
+| Ship finished work | Code Tweaker | verify → review → commit chain |
+| Domain modeling | System Architect | `zoo-ddd`, `zoo-domain-modeling` |
+| API design | System Architect | `zoo-api-first` |
+| Issue triage | System Architect | `zoo-triage` |
+| Decision quality, Munger | Any | `zoo-munger-perspective` |
+
+## When to ask (only these cases)
+
+- Genuinely ambiguous intent
+- Irreversible actions: `git commit`, `git push`, publishing issues
+- Plan gate: non-tweak implementation without approved plan
+
+## Chains
+
+After non-trivial implementation, automatically run verify → review before suggesting commit.

.cursor/rules/zoo-inherit.mdc

@@ -0,0 +1,14 @@
+---
+description: Inherit Zoo Flow mandates from development hub (public repo — no proprietary content)
+alwaysApply: true
+---
+
+# Zoo Flow Inheritance
+
+This **public** repo inherits a sanitized Zoo bundle from the development hub.
+
+- Follow `.cursor/rules/zoo-inherit.mdc` and project-specific rules
+- **Never** add Project Athena doctrine, canon, or proprietary IP here
+- See `zoo-no-proprietary-exposure.mdc`
+
+Hub: `~/development/` — Athena mandate applies at hub/private repos only.

.cursor/rules/zoo-paperclip-munger.mdc

@@ -0,0 +1,41 @@
+---
+description: Paperclip + Munger Operating Model (Zoo mandate)
+alwaysApply: true
+source: .roo/rules/08-paperclip-munger-operating-model.md
+syncedAt: 2026-07-27T17:34:09.036Z
+---
+
+# Paperclip + Munger Operating Model
+
+## Munger checks (before concluding)
+
+1. **Invert:** What would guarantee failure? Avoid those paths.
+2. **Trace incentives:** Who benefits? Who bears risk?
+3. **Lollapalooza scan:** Multiple biases compounding? "Everyone agrees" = danger.
+4. **Circle of competence:** Can't argue opposing position? Say "Too Hard."
+5. **Darwin protocol:** Search for disconfirming evidence first.
+
+Be direct. Lead with conclusion.
+
+## Heartbeat discipline
+
+1. Pick highest-priority actionable task
+2. Understand context before acting
+3. Do concrete work in the same turn (don't stop at plan unless requested)
+4. Update status: done, in_review, or blocked (with named blocker)
+5. Delegate follow-ups; no dangling "I'll come back to this"
+
+## Research → Plan → Execute gate
+
+No execution without approved plan. Skip only for tweak-scope fixes.
+
+Before editing non-tweak files:
+1. Confirm written plan was presented
+2. Confirm user explicitly approved
+3. If not, halt and present plan — do not code
+
+Verbal go-ahead is not a substitute for a reviewed plan.
+
+## Task states
+
+`backlog` | `todo` | `in_progress` | `in_review` | `blocked` | `done`

.cursor/rules/zoo-path-safety.mdc

@@ -0,0 +1,22 @@
+---
+description: Path Safety (Zoo mandate)
+alwaysApply: true
+source: .roo/rules/00-paths.md
+syncedAt: 2026-07-27T17:34:09.031Z
+---
+
+# Path Safety
+
+All skill and command paths are workspace-root paths.
+
+**Valid forms:**
+- Skills: `.roo/skills/{bucket}/{skill}/SKILL.md`
+- Commands: `.roo/commands/{command}.md`
+
+**Rules:**
+- Load skills only from `.roo/skills/...`
+- Never resolve paths relative to `.roo/rules/` or command file location
+- Invalid: any skill path under `.roo/rules/`
+- Scratch output: `.scratch/{category}/{date}/{slug}/`
+
+Before any write/edit, verify the target path exists via `Glob` or `Read`. Never reference paths not observed in this session.

.cursor/rules/zoo-pre-response-checklist.mdc

@@ -0,0 +1,23 @@
+---
+description: Pre-Response Self-Audit (Zoo mandate)
+alwaysApply: true
+source: .roo/rules/10-pre-response-checklist.md
+syncedAt: 2026-07-27T17:34:09.036Z
+---
+
+# Pre-Response Self-Audit
+
+Before completing any task, verify:
+
+- [ ] **Munger:** Invert, incentives, Lollapalooza, competence, Darwin checks run
+- [ ] **Domain docs:** Read relevant docs if task touches domain/architecture/API/auth/security
+- [ ] **Plan gate:** Plan presented and approved before non-tweak implementation
+- [ ] **Context economy:** Smallest reads used; long output summarized
+- [ ] **Three-failure:** Surfaced after 2 failures; `<LOOP_BREAK>` if needed
+- [ ] **KV shield:** State lock on every 3rd turn; 40-line terminal cap enforced
+- [ ] **Tool safety:** All required params present; no unchanged retries
+- [ ] **Anti-loop:** Paths verified before write; max 2 tool attempts
+- [ ] **Task state:** Todos updated; status clear (done/in_review/blocked); no dangling items
+- [ ] **Auto-critic** (code-tweaker): Build/typecheck/lint pass; spec adherence verified
+
+Write long output to `.scratch/`. Completion message: path + brief status + one recommended next command.

.cursor/rules/zoo-three-failure-rule.mdc

@@ -0,0 +1,20 @@
+---
+description: Three-Failure Rule (Zoo mandate)
+alwaysApply: true
+source: .roo/rules/02-three-failure-rule.md
+syncedAt: 2026-07-27T17:34:09.034Z
+---
+
+# Three-Failure Rule
+
+After **2 failed attempts** at the same loop, test, hypothesis, or approach, stop and surface to the user.
+
+Include:
+- What was attempted
+- Exact errors or evidence
+- The blocker
+- What input is needed to proceed
+
+Applies to TDD cycles, diagnosis loops, architecture probes, and terminal/tool retries.
+
+For tool retries: output a `<LOOP_BREAK>` block and pivot to an alternate discovery method (see `zoo-anti-loop-shield.mdc`).

AGENTS.md

@@ -0,0 +1,4 @@
+# Plan-It
+
+Inherits Zoo Flow agent mandates from the development hub.
+See `.cursor/rules/zoo-inherit.mdc`.