--- name: pws-brain description: "RETIRED TEST HARNESS (superseded 2026-07-22): compared the then-production mindrian-brain MCP against the neo4j-agent Aura Agent over Neo4j. Both routes are superseded by the unified pws-brain-mcp Memgraph backend. Kept for deliberate historical re-runs only." help_jtbd: "Ask one generic methodology question and see how the production Brain and the experimental Aura Agent each answer it, side by side." body_shape: E (Action Report) layer: "none" layer_why: "Runs the same methodology question against the experimental Aura Agent next to the production Brain and shows both; a diagnostic comparison view." body_shape_detail: two labeled answer blocks (A production Brain, B Aura Agent) plus a short observed-differences note hitl_shape: "none" hitl_why: "A read-only comparison report that runs one question through two backends and takes no navigator decision, so it reaches no genuine fork." argument-hint: "" serves_jtbd: ["audit-room"] teaching: "When you want to see how the experimental Aura Agent answers next to the production Brain, /mos:pws-brain runs the same methodology question through both and shows the answers side by side. An evaluation harness, not a production surface." disable-model-invocation: true # Reward-before-investment (docs/reward-before-investment-rule.md): explicit opt-out. # This is a RETIRED, deliberately-invoked evaluation harness, not a navigator entry # flow, so it has no first variable reward to deliver and never will. interactive_first_reward: --none (scripting only) ui_reference: skills/ui-system/SKILL.md allowed-tools: - mcp__mindrian-brain__brain_ask - mcp__mindrian-brain__brain_search - mcp__mindrian-brain__brain_query - mcp__neo4j-agent__Larry # --- Quick-260716-VFT CIRS R1 exclude (Canon Part 11) --- # Born WIRED-or-EXCLUDED: this surface is declared EXCLUDED-with-reason, not left # dark. It is a deliberately-run evaluation harness (mirrors /mos:agentshield's # precedent), so it never reacts to a navigator problem-state and takes no reach # on the spine. connector: excluded: true reason: "RETIRED evaluation harness (declared 2026-07-16, superseded 2026-07-22). It compared the then-production mindrian-brain backend against the neo4j-agent Aura Agent candidate over the same live Neo4j graph. BOTH routes are now superseded by the unified pws-brain-mcp Memgraph backend, whose origin lib/core/brain-client.cjs defines once as its BRAIN_URL default (Phase 339 moves that origin from Memgraph to Theo; this reason string does not hardcode a host so it does not need updating again when it does). Kept on disk as historical reference for a deliberate re-run; fires no reach, opens no spine wire, never sensor-triggered (Part 11 R1 EXCLUDED-with-reason, mirrors /mos:agentshield)." --- # /mos:pws-brain > **RETIRED / STALE as of 2026-07-22.** Do NOT run this to learn how the Brain > behaves today. BOTH routes compared below are superseded: route A's > `mindrian-brain` MCP and route B's `neo4j-agent` Aura Agent both spoke to the old > Neo4j graph. The live backend is now the unified `pws-brain-mcp` Memgraph service. > In-repo proof the migration already shipped: `lib/core/brain-client.cjs` names > that service's origin as its `BRAIN_URL` default, and its file header names it > "step 4 of the 2026-07-22 Memgraph migration" (check that file, not this doc, for > the current URL -- Phase 339 is moving it again, to Theo). A run of this harness > therefore measures a backend pair production no longer uses. Everything below is > preserved as historical reference for a deliberate re-run, not as live truth. **EXPERIMENTAL.** This is a side-by-side evaluation harness. It runs the SAME methodology question through two different routes to the SAME live Neo4j teaching graph (Instance01, "PWS Framework - Mondrian") and shows both answers next to each other. - **Route A - the production Brain.** The shipped `mindrian-brain` MCP (Neo4j plus Pinecone vectors). This is the real Brain every user reaches. - **Route B - the Aura Agent (internally named Larry).** The `neo4j-agent` MCP, one tool with five Aura sub-tools: `classify_problem_type`, `find_frameworks_for_problem_type`, `Text2Cypher`, `Search` over `MethodologyChunk` embeddings, and `find_commands_for_problem_type`. This command exists to EVALUATE route B as a candidate alternative path. It is NOT a production reach, it does NOT replace any Brain path, and it is never sensor-triggered. The navigator invokes it deliberately or not at all. ## Part 8 Boundary (LOCKED) BOTH backends may ONLY receive GENERIC methodology queries: framework names, problem-type classifications, command names, and public topic terms. NEVER room content, user artifacts, venture specifics, meeting text, or personal data. This rule binds route B exactly as hard as route A. The Aura Agent is the same graph behind a different door, and an evaluation harness gets NO boundary discount. If the navigator's question carries room-specific or venture-specific material, do NOT send it to either backend. Stop and ask for a generic reformulation, and offer an example of how to genericize: > \- Instead of naming the venture, ask the shape of the problem. > \- Bad (venture-specific): "which frameworks fit Acme's B2B pricing rollout?" > \- Good (generic): "which frameworks fit a market-sizing problem?" Only once the question is generic does the run proceed. ## Pre-flight Two availability checks. Either backend missing means STOP: a one-sided run is not a comparison. Never silently degrade to a single backend and present it as a comparison result. **Check 1 - production Brain.** If the `mcp__mindrian-brain__*` tools are unavailable: ``` x Production Brain not connected Why: the mindrian-brain MCP server is not reachable in this session Fix: connect the Brain MCP, then re-run /mos:pws-brain ``` **Check 2 - Aura Agent.** If `mcp__neo4j-agent__Larry` is unavailable: ``` x Aura Agent not available Why: this command needs the USER-LEVEL neo4j-agent MCP server, a dev-machine evaluation dependency that does NOT ship with the plugin Fix: add or enable the neo4j-agent server in your Claude Code MCP config, then re-run /mos:pws-brain ``` Both present -> proceed. Either absent -> stop with the matching error above. ## Run the comparison Take `$ARGUMENTS` as the methodology question. If it is empty, ask for one generic methodology question and stop. Run the Part 8 screen above FIRST. Only after it passes, send the SAME question text to both backends: - **Route A** calls `mcp__mindrian-brain__brain_ask` with the question. If `brain_ask` errors, fall back ONCE to `mcp__mindrian-brain__brain_search` and LABEL the fallback in the output. - **Route B** calls `mcp__neo4j-agent__Larry` with the IDENTICAL string in its query parameter, verbatim, no rephrasing. Rephrasing would invalidate the comparison. Call A, then B. Do NOT merge, summarize, or cross-pollinate one answer into the other's call. Each backend answers the same question cold. ## Present the report (Shape E) Render one action report: ``` ------------------------------ Comparison: "[the question asked]" ------------------------------ A. mindrian-brain (production) [route A's answer verbatim; note "(brain_search fallback)" if it fell back] B. neo4j-agent Aura Agent (experimental) [route B's answer verbatim] Observed differences \- Coverage: [which answer named more, or fewer, frameworks/commands] \- Grounding: [does each answer cite frameworks/commands that exist in the graph] \- Specificity: [which answer was more concrete vs more generic] \- Refusal / deferral: [did either backend refuse, hedge, or defer] \- [one more only if there is a real fifth difference] ------------------------------ ``` Never declare a winner. Never grade. The navigator judges. State the observed differences plainly and stop. ## Zone 4 (Action Footer) > \- Note what you saw for the ongoing Aura Agent evaluation. > \- Run another generic question -> /mos:pws-brain ""