docs: PicoClaw fact-check tool order before a factual reply (#22)

search then get then audit. throttled is not absence. 2dph is the
gate, not the agent loop.

AGENTS.md

@@ -15,6 +15,9 @@ Read first: [PLAN](PLAN.md) → [docs](docs/).
 - `info` root = descriptive/narrative leafs, searchable, never asserted as fact.
 - Search is deduction: `facts` → `info` → `web-search` (second independent
   source). An answer is `confirmed` only if it comes off the facts root.
+- Fact-check every *claim* (facts → info → live → web), not every edit or
+  syntax tweak. PicoClaw: `search` then `get` then `audit` before a factual
+  reply (`skills/picoclaw/SKILL.md`). `throttled` is not a negative finding.
 
 ## Hard rules
 

PLAN.md

@@ -51,7 +51,7 @@ detective method: **a fact needs ≥2 independent sources or it is
 2dph/
   PLAN.md / AGENTS.md
   docs/                     published docs (this conversation → docs/ as md)
-  skills/                   in-project skills (web-search, postgres, brain, diataxis-docs)
+  skills/                   in-project skills (web-search, postgres, brain, picoclaw, diataxis-docs)
   bin/
     facts/extract.go audit.go crm.go  # D14 shebang; Python implementation
     kb/index                Python write path (called by bin/brain/index.go)

bin/tools/test_skills.py

@@ -37,3 +37,13 @@ class SkillsTest(unittest.TestCase):
         self.assertIn("tools.md", skill)
         for name in ("search", "get", "stats", "audit"):
             self.assertIn(f"`{name}`", tools)
+
+    def test_picoclaw_lists_tool_order(self) -> None:
+        skill = (ROOT / "skills" / "picoclaw" / "SKILL.md").read_text()
+        agents = (ROOT / "AGENTS.md").read_text()
+        self.assertIn("**`search`**", skill)
+        self.assertIn("**`get`**", skill)
+        self.assertIn("**`audit`**", skill)
+        self.assertIn("throttled", skill.lower())
+        self.assertIn("not a negative finding", agents)
+        self.assertIn("Fact-check every", agents)

skills/picoclaw/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: picoclaw
+description: >-
+  2dph is the memory/fact gate, not the agent loop. Use when wiring PicoClaw
+  or any MCP client: call brain search/get/audit before a factual reply.
+  throttled is not a negative finding.
+---
+
+# PicoClaw — fact-check before assert
+
+PicoClaw (or any agent) speaks MCP at `POST /mcp` on `bin/brain/serve.go`.
+2dph does not run the agent loop. Compose profile for a live agent is issue #9.
+
+## Tool order (before a factual reply)
+
+1. **`search`** — facts root first, then info. The `web` block is a second
+   source when there is no facts hit. Status `throttled` / `skipped` /
+   `refused` is **not** evidence of absence.
+2. **`get`** — full leaf body only when a hit `id` is needed.
+3. **`audit`** — if recall or confidence looks wrong.
+
+Then answer. Confirmed only from facts (≥2 independent sources). Anything
+else is `(not confirmed)`. Missing graph ≠ “does not exist”.
+
+Generated tool list: [../brain/tools.md](../brain/tools.md).