@dzhechkov/harness-cli 0.3.191 → 0.3.193

package/README.md

@@ -363,7 +363,7 @@ dz setup             --target <name> [--preset <name>] [--select id,id,...] [--s
 dz init              --target <name> [--preset <name>] [--select id,id,...] [--force]
 dz install           <npm-pkg> [--target <name>] [--project <dir>] [--force]
 dz bundle            [--preset <name> | --select id,...] [--out <dir>] [--skills-dir <dir>] [--force]
-dz teach             "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--no-mirror]
+dz teach             "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--project <dir>] [--no-mirror]   # --project pins the learned store to <dir>/.dz (not the cwd) — pin to a canonical brain
 dz teach --from-json <file> [--project <dir>] [--no-mirror]   # bulk-import a `dz recall --all --json` export (share a learned store across machines)
 dz consolidate       [--sessions-dir <dir>] [--project <dir>] [--no-mirror]
 dz recall            "<query>" [--limit <N>] [--semantic | --no-semantic] [--project <dir>]   # hybrid lexical+vector when the vector tier is enabled
@@ -1125,6 +1125,30 @@ dz vector harmonize --apply                   #  collapse (backup written first;
 dz teach --harmonize                          #  same thing — documented alias
 ```
 
+### One canonical brain store — sharing + fragmentation recovery
+
+The self-learning loop is only compounding if every run reads and writes the **same** store. Two things make that hold:
+
+- **One brain grows across runs.** The `feature-adr` workflow pins its Step-0 `dz recall` and Step-8 `dz teach` to `args.brain` (default = the workspace root), so every run reads and writes the *same* `<brain>/.dz` store and the pool count in `dz statusline` climbs monotonically. Because `dz teach --project <dir>` writes to `<dir>/.dz` (not the process cwd), the lesson lands in the brain even when a code agent has `cd`'d into a target repo. Override `args.brain` with a stable absolute path to keep one brain across several target checkouts.
+- **Share with colleagues (idempotent):**
+  ```bash
+  dz recall --all --json > patterns.json                 #  export the WHOLE store (SHARING form)
+  dz teach --from-json patterns.json --project <brain>   #  import on their machine (exact-text dedup)
+  ```
+- **Recover an accidentally-fragmented store.** If a lesson was taught while cd'd into a stray repo (so it landed in `<stray>/.dz`), merge it back into the canonical brain — a JSON round-trip, no new tooling, dedup-by-exact-text so it is **idempotent** (re-running imports zero):
+  ```bash
+  cd <stray-repo> && dz recall --all --json > /tmp/stray.json   #  export the stray store
+  dz teach --from-json /tmp/stray.json --project <brain>        #  merge into the canonical brain
+  dz vector harmonize --apply --project <brain>                 #  optional: collapse near-dup paraphrases
+  ```
+
+- **Route each stage to an optimal model — `args.models`.** The `feature-adr` workflow takes an optional per-stage map (`{router, requirements, research, adr, ideation, ddd, architecture, plan, code, qe, fleet}`); each value a **spec** — Claude `fable|opus|sonnet|haiku` or Codex `codex[:id[:reasoning]]` (ids incl. `gpt-5.5`, `gpt-5.6`) — backed by a proven default table. The **load-bearing default**: when `models.qe` is unset, QE auto-routes to the *other family than the coder* (a model that writes code must not also self-QE — cross-model review catches what self-review misses), falling back to a Claude reviewer if Codex is absent (never blocks). `args.models[stage]` wins over the legacy `planner`/`coder`/`qeReviewer`/`codexModel` knobs; omitting it is byte-identical to today; the result reports `modelsUsed`. Adding a new Codex id is a data-only allowlist edit (gpt-5.6-ready).
+  ```js
+  // Claude codes, Codex independently QEs (cross-model by construction):
+  Workflow({ scriptPath: '.claude/workflows/feature-adr.js',
+    args: { slug: 'add-oauth', description: '…', tier: 'L', models: { code: 'opus', qe: 'codex:gpt-5.6:high' } } })
+  ```
+
 **"Verify the whole thing is healthy."**
 ```bash
 dz doctor                                        # node, skills, backend

package/dist/cli.js

@@ -30,7 +30,7 @@ Usage:
   dz bundle  [--preset <name> | --select id,id,...] [--out <dir>] [--skills-dir <dir>] [--force]   (portable self-contained skill bundles for a generic/LangGraph consumer)
   dz publish [--filter <name>] [--bump-only]   (dry-run by default; pass --yes/--confirm/--no-dry-run to go live)
   dz setup --target <name> [--preset <name>] [--select id,id,...] [--skills-dir <dir>] [--project <dir>] [--memory agentdb] [--no-memory] [--no-hooks] [--install-driver] [--force] [--enrich]
-  dz teach "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--no-mirror]
+  dz teach "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--project <dir>] [--no-mirror]   (--project pins the learned store to <dir>/.dz, not the cwd — pin to a canonical brain)
   dz teach --from-json <file> [--project <dir>] [--no-mirror]   (bulk-import a 'dz recall --all --json' export — share a learned store across machines)
   dz consolidate [--sessions-dir <dir>] [--project <dir>] [--no-mirror] [--prune-noise]
   dz recall "<query>" [--limit <N>] [--semantic | --no-semantic] [--books [--book <slug>]] [--project <dir>] | dz recall --all [--json]

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dzhechkov/harness-cli",
-  "version": "0.3.191",
+  "version": "0.3.193",
   "description": "The dz CLI — install AI skills for Claude Code, Codex, OpenCode, Hermes, OpenClaude, GitHub Copilot. 35 commands, 13 presets, 6 platform targets.",
   "type": "module",
   "license": "MIT",
@@ -54,7 +54,7 @@
     "@dzhechkov/skills-reverse-engineering": "^0.1.0",
     "@dzhechkov/skills-presentation-storyteller": "^0.1.0",
     "@dzhechkov/skills-website-cloner": "^0.1.0",
-    "@dzhechkov/harness-core": "0.3.88"
+    "@dzhechkov/harness-core": "0.3.90"
   },
   "devDependencies": {
     "@types/node": "^25.6.0",

package/src/cli.ts

@@ -100,7 +100,7 @@ Usage:
   dz bundle  [--preset <name> | --select id,id,...] [--out <dir>] [--skills-dir <dir>] [--force]   (portable self-contained skill bundles for a generic/LangGraph consumer)
   dz publish [--filter <name>] [--bump-only]   (dry-run by default; pass --yes/--confirm/--no-dry-run to go live)
   dz setup --target <name> [--preset <name>] [--select id,id,...] [--skills-dir <dir>] [--project <dir>] [--memory agentdb] [--no-memory] [--no-hooks] [--install-driver] [--force] [--enrich]
-  dz teach "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--no-mirror]
+  dz teach "<pattern>" [--reward <0-1>] [--domain <name>] [--type rule|success-pattern|lesson-learned] [--project <dir>] [--no-mirror]   (--project pins the learned store to <dir>/.dz, not the cwd — pin to a canonical brain)
   dz teach --from-json <file> [--project <dir>] [--no-mirror]   (bulk-import a 'dz recall --all --json' export — share a learned store across machines)
   dz consolidate [--sessions-dir <dir>] [--project <dir>] [--no-mirror] [--prune-noise]
   dz recall "<query>" [--limit <N>] [--semantic | --no-semantic] [--books [--book <slug>]] [--project <dir>] | dz recall --all [--json]