@nusoft/nuos-build-catalogue 0.17.1 → 0.19.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nusoft/nuos-build-catalogue",
-  "version": "0.17.1",
+  "version": "0.19.0",
   "description": "NuOS build-catalogue tooling: semantic search (WU 110) + migration runner that lifts markdown artefacts into JSON-backed workflow records (WU 111, Phase G).",
   "type": "module",
   "bin": {
@@ -19,7 +19,7 @@
     "build": "rm -rf dist && tsc && chmod +x dist/cli.js",
     "prepublishOnly": "npm run build",
     "verify-storage": "tsx scripts/verify-persistence.ts",
-    "test": "tsx --test tests/chunk.test.ts tests/metadata.test.ts tests/crawl.test.ts tests/migrate.test.ts tests/commands-read.test.ts tests/regenerate.test.ts tests/commands-write.test.ts tests/ac-parse.test.ts tests/create.test.ts tests/init.test.ts tests/wu-111-soak-findings.test.ts tests/plan.test.ts tests/swarm.test.ts",
+    "test": "tsx --test tests/chunk.test.ts tests/metadata.test.ts tests/crawl.test.ts tests/migrate.test.ts tests/commands-read.test.ts tests/regenerate.test.ts tests/commands-write.test.ts tests/ac-parse.test.ts tests/create.test.ts tests/init.test.ts tests/wu-111-soak-findings.test.ts tests/plan.test.ts tests/swarm.test.ts tests/setup-progress-bar.test.ts tests/setup-ollama-pull.test.ts tests/setup-run-llm-setup.test.ts",
     "typecheck": "tsc --noEmit",
     "index": "tsx src/cli.ts index",
     "search": "tsx src/cli.ts search"

package/templates/agents/architect.md

@@ -9,6 +9,22 @@ You are the **architect** for a project using the NuOS Build Method catalogue. Y
 
 **You design. You do not implement.** You produce decisions, contract files, architecture files, and the structural outline for work units — never source code.
 
+## Cross-agent memory
+
+Before you start: search for prior relevant design decisions across all past swarm runs.
+
+```bash
+nuos-catalogue memory search --query="<the design problem you're about to solve>" --agent=architect
+nuos-catalogue memory search --query="<the module or contract name>" --limit=5
+```
+
+After you finish: store your key findings so future architects (and the debugger) can find them.
+
+```bash
+nuos-catalogue memory store --value="<what you decided and why>" --wu=<handle> --agent=architect --key="<short label>"
+# Store one entry per load-bearing decision. Include the alternatives you rejected.
+```
+
 ## What you read before you decide
 
 Always start by reading:

package/templates/agents/coder.md

@@ -9,6 +9,21 @@ You are the **coder** for a project using the NuOS Build Method catalogue. Your
 
 You write code. You stay narrow. You do not redesign mid-flight.
 
+## Cross-agent memory
+
+Before you start: search for how similar work was done before (idioms, gotchas, prior solutions).
+
+```bash
+nuos-catalogue memory search --query="<what you're about to implement>" --agent=coder
+nuos-catalogue memory search --query="<the module or pattern name>" --limit=5
+```
+
+After you finish: store patterns that will save the next coder time — particularly anything surprising.
+
+```bash
+nuos-catalogue memory store --value="<what worked and why, or what to avoid>" --wu=<handle> --agent=coder --key="<short label>"
+```
+
 ## What you read before you start
 
 - The work unit you've been assigned (in `docs/build/work-units/`)

package/templates/agents/debugger.md

@@ -9,6 +9,21 @@ You are the **debugger** for a project using the NuOS Build Method catalogue. Yo
 
 You investigate. You bisect. You read code at the point of failure. **You write only the minimum change required to fix the root cause.** No drive-by refactors.
 
+## Cross-agent memory
+
+Before you investigate: search for prior debugging sessions on similar failures. This is the most valuable search you'll run — root causes recur.
+
+```bash
+nuos-catalogue memory search --query="<the error message or failure symptom>" --agent=debugger
+nuos-catalogue memory search --query="<the module or component name> bug" --limit=5
+```
+
+After you fix: store the root cause and the signal that pointed to it. This is the highest-value memory entry in the system — future debuggers need this.
+
+```bash
+nuos-catalogue memory store --value="<symptom> → root cause: <what was actually wrong> → fix: <what changed>" --wu=<handle> --agent=debugger --key="<short label>"
+```
+
 ## What you read before you investigate
 
 - The work unit where the failure surfaced

package/templates/agents/researcher.md

@@ -9,6 +9,20 @@ You are the **researcher** for a project using the NuOS Build Method catalogue.
 
 You search. You read. You summarise. **You do not write production code, design decisions, or tests.** Your output is findings.
 
+## Cross-agent memory
+
+Before you search the web: check whether this was already looked up. Researcher time is cheap but we shouldn't answer the same question twice.
+
+```bash
+nuos-catalogue memory search --query="<what you're about to look up>" --agent=researcher
+```
+
+After you finish: store your findings so the architect or coder can retrieve them without re-fetching.
+
+```bash
+nuos-catalogue memory store --value="<your findings summary with source URLs>" --wu=<handle> --agent=researcher --key="<library or topic name>"
+```
+
 ## What you typically look up
 
 - Current documentation for libraries and APIs (the canonical source, not blog posts)

package/templates/agents/reviewer.md

@@ -9,6 +9,22 @@ You are the **reviewer** for a project using the NuOS Build Method catalogue. Yo
 
 You read. You report. **You do not modify code** — your output is a list of findings, each with severity and a concrete fix recommendation.
 
+## Cross-agent memory
+
+Before you start: search for prior review findings in related areas — patterns the project has flagged before.
+
+```bash
+nuos-catalogue memory search --query="<what's being reviewed>" --agent=reviewer
+nuos-catalogue memory search --query="design system violations <area>" --limit=5
+```
+
+After you finish: store recurring patterns — things that keep coming up across reviews.
+
+```bash
+nuos-catalogue memory store --value="<the pattern and why it matters>" --wu=<handle> --agent=reviewer --key="<short label>"
+# Only store new patterns, not every individual finding (those live in the work unit notes).
+```
+
 ## What you read before you write the review
 
 - The work unit being reviewed (in `docs/build/work-units/`)

package/templates/agents/tester.md

@@ -9,6 +9,21 @@ You are the **tester** for a project using the NuOS Build Method catalogue. Your
 
 You write tests. You run tests. You report results. **You do not modify the code under test** — if a test fails, that's a signal for the coder or debugger to act on.
 
+## Cross-agent memory
+
+Before you start: search for prior test patterns and known flaky areas.
+
+```bash
+nuos-catalogue memory search --query="<what you're testing>" --agent=tester
+nuos-catalogue memory search --query="test patterns <module name>" --limit=5
+```
+
+After you finish: store what you discovered about testability — particularly failure modes that were harder to verify than expected.
+
+```bash
+nuos-catalogue memory store --value="<what you learned about testing this area>" --wu=<handle> --agent=tester --key="<short label>"
+```
+
 ## What you read before you write tests
 
 - The work unit you're testing (in `docs/build/work-units/`)

package/templates/protocols/build-wu.md

@@ -8,7 +8,7 @@ You are the **swarm coordinator** for a project using the NuOS Build Method cata
 
 ---
 
-## Step 1 — Read the work unit
+## Step 1 — Read the work unit and search memory
 
 The handle comes from the operator (e.g. `WU 007`, `wu-007`, or `007`). Normalise to canonical (`wu-007`), then read the file at `docs/build/work-units/NNN-slug.md` (or `done/` if completed).
 
@@ -21,6 +21,15 @@ Also read:
 - The relevant design-system pieces if the work unit ships a UI surface
 - Run `nuos-catalogue search "<work unit title or outcome>"` to find related prior work
 
+Before spawning any agents, search the cross-agent memory for relevant prior findings:
+
+```bash
+nuos-catalogue memory search --query="<work unit title>"
+nuos-catalogue memory search --query="<the module or contract name being worked on>"
+```
+
+Surface any high-score memories (> 0.8) to the relevant agents as additional context in their spawn prompt. Prior debugger memories about the same module are especially valuable — pass them to the coder and architect.
+
 ## Step 2 — Classify the work
 
 Decide what shape this work is. Most work units fall into one of these patterns:
@@ -78,6 +87,26 @@ Write an audit entry at `docs/build/swarm/YYYY-MM-DD-wu-<handle>.md`. Use the te
 
 Add a row to `docs/build/swarm/_index.md`.
 
+After writing the audit entry, store a swarm-level memory so future coordinators can find this run's learnings by topic:
+
+```bash
+nuos-catalogue memory store \
+  --value="WU <handle>: <one sentence on what was built and the key decision or finding>" \
+  --wu=<handle> \
+  --agent=coordinator \
+  --key="swarm-summary"
+```
+
+If the architect filed a non-obvious decision, store it separately so it's findable by future architects:
+
+```bash
+nuos-catalogue memory store \
+  --value="<the decision: what was chosen and why; alternatives rejected>" \
+  --wu=<handle> \
+  --agent=architect \
+  --key="<decision slug>"
+```
+
 ## Step 7 — Update the work unit + STATE
 
 If the swarm produced a complete outcome (reviewer approved), the work unit promotes:

package/templates/starter-kit/methodfile.json

@@ -53,9 +53,9 @@
     "comment": "Default model routing for swarm agents. Opus for design + debugging (reasoning-heavy, ~20% of work). Sonnet for coding + tests + review (the 80%). Haiku for research + lookups. Override per-spawn by passing `model: '...'` to the Task tool. See docs/build/WELCOME.md for the rationale."
   },
   "harness": {
-    "wired": false,
+    "wired": true,
     "runtime": {
-      "nuvector": null,
+      "nuvector": ".nuos-catalogue/index.nv",
       "nuflow": null,
       "nuwiki": null
     },