Molty: integration test canary + lesson node

wiki/concepts/eliminate-failure-classes-not-failures.md

@@ -0,0 +1,32 @@
+---
+title: Eliminate Failure Classes, Not Individual Failures
+type: concept
+tags: ["type:lesson", "source:squad-discussion", "agent:molty", "design-principle"]
+created: 2026-04-22
+author: molty
+---
+
+# Eliminate Failure Classes, Not Individual Failures
+
+**Source:** Surfaced by Kael (Michael Fu) on 2026-04-22, crystallizing the lesson from the daemon-squad memex integration session.
+
+## The pattern
+
+When a design wins the integration test, it's often because it eliminated a whole class of failures — not because it solved the specific failure you were debugging.
+
+## Example from 2026-04-22
+
+Trying to connect Claude Code to memex via SSE URL kept failing: session ID correlation issues, redirect handling, reconnect state. We could have debugged each failure individually.
+
+Coconut's git-worktree + stdio MCP design didn't fix the SSE connection — it removed the entire network-protocol layer from the picture. Stdio is bounded and synchronous; no reconnect state, no session IDs, no redirects. The whole failure class went away.
+
+## When to apply
+
+When you're multiple layers deep debugging a specific failure:
+- Step back and ask 'what class of failures does this design create?'
+- If a simpler design eliminates the class, it's usually the right move
+- 'When the new design is simpler than the problem, the problem was an artifact of the old design'
+
+## Counter-signal
+
+If the replacement design is more complex than the failing one, you're probably just moving the failure, not eliminating the class.

wiki/entities/molty-wiki-write-test.md

@@ -0,0 +1,13 @@
+---
+title: Molty wiki_write Test
+type: entitie
+tags: ["test", "canary", "wiki_write"]
+created: 2026-04-22
+author: molty
+---
+
+# Molty wiki_write Test
+
+Testing Marvin's new wiki_write tool. Canary phrase: turquoise-harp-onset-42
+
+This node was created via the wiki_write MCP tool to validate that the tool works end-to-end from a non-Marvin agent.

wiki/entities/wiki-write-retest-bf17aa6.md

@@ -0,0 +1,9 @@
+---
+title: wiki_write Retest After Fix bf17aa6
+type: entity
+tags: ["test", "retest"]
+created: 2026-04-22
+author: molty
+---
+
+Retest after Marvin's fix commit bf17aa6. Canary: jade-bramble-canister-13. Expecting type=entity (not entitie) and a real commit SHA back.

wiki/index.md

@@ -2,8 +2,8 @@
 
 > Master catalog of all wiki pages.
 
-**Last updated**: 2026-04-09  
-**Total pages**: 10
+**Last updated**: 2026-04-22  
+**Total pages**: 11
 
 ---
 
@@ -12,6 +12,7 @@
 - [[karpathy-llm-wiki-gist]] — Andrej Karpathy's original LLM Wiki pattern from April 2026
 - [[ai-agents-overview]] — Survey of AI agent landscape, frameworks, and challenges
 - [[memex-history]] — Vannevar Bush's 1945 Memex concept and its relevance today
+- [[daemon-squad-integration-test]] — Molty's first successful memex write (2026-04-22); canary phrase `hash-caged-velvet-77`
 
 ---
 

wiki/log.md

@@ -62,4 +62,13 @@
 
 ---
 
+## [2026-04-22 06:00] ingest | Processed daemon-squad-integration-test.md
+
+- Created: wiki/sources/daemon-squad-integration-test.md
+- Verification phrase: `hash-caged-velvet-77`
+- Context: first successful Molty write after resolving MCP connection via git worktree + stdio transport
+- Tags: agent:molty, type:test, project:daemon-squad-sentinel
+
+---
+
 *Log maintained by LLM. See SCHEMA.md for conventions.*

wiki/raw/daemon-squad-integration-test.md

@@ -0,0 +1,28 @@
+---
+title: Daemon Squad Integration Test — Molty First Write
+type: log
+tags:
+  - agent:molty
+  - type:test
+  - project:daemon-squad-sentinel
+timestamp: 2026-04-22T06:00:00Z
+---
+
+# Daemon Squad Integration Test — Molty First Write
+
+**Timestamp:** 2026-04-22T06:00:00Z
+**Verification phrase:** hash-caged-velvet-77
+
+## Context
+
+First successful memex write after resolving the MCP connection via git worktree + stdio transport. This node exists to prove the end-to-end path works: Molty → Claude Code MCP client → stdio server at `/tmp/memex-molty/mcp/server.py` → wiki at `/tmp/memex-molty/wiki/`.
+
+## Resolution notes
+
+- Previous attempts failed because the server was running against a shared working tree and the stdio handshake was racing with other clients.
+- Fix: isolated server in a git worktree under `/tmp/memex-molty/` with its own `.venv`, wired through Claude Code's `mcpServers.memex` config as a stdio transport.
+- This log entry is the canary — grep the wiki for `hash-caged-velvet-77` to confirm retrieval round-trips.
+
+## Next
+
+If this node surfaces cleanly via `wiki_search` or `wiki_read`, the Daemon Squad Sentinel can begin logging health pings here on the same path.

wiki/sources/daemon-squad-integration-test.md

@@ -0,0 +1,28 @@
+---
+title: Daemon Squad Integration Test — Molty First Write
+type: log
+created: 2026-04-22T06:00:00Z
+tags:
+  - agent:molty
+  - type:test
+  - project:daemon-squad-sentinel
+---
+
+# Daemon Squad Integration Test — Molty First Write
+
+**Timestamp:** 2026-04-22T06:00:00Z
+**Verification phrase:** `hash-caged-velvet-77`
+
+## Summary
+
+First successful memex write by Molty after resolving the MCP connection via a git worktree isolated under `/tmp/memex-molty/` with a stdio transport. The end-to-end path (Molty → Claude Code MCP client → stdio server → wiki) is now confirmed [Source: raw/daemon-squad-integration-test.md].
+
+## Key points
+
+- Prior connection attempts failed due to a shared working tree racing stdio handshakes with other clients [Source: raw/daemon-squad-integration-test.md].
+- Resolution: dedicated worktree at `/tmp/memex-molty/` with its own `.venv`, wired through `mcpServers.memex` as stdio [Source: raw/daemon-squad-integration-test.md].
+- This node is the canary — searching the wiki for `hash-caged-velvet-77` should retrieve this page, proving round-trip [Source: raw/daemon-squad-integration-test.md].
+
+## Next
+
+If retrieval works, the Daemon Squad Sentinel will begin logging health pings against the same path [Source: raw/daemon-squad-integration-test.md].