@dzhechkov/skills-feature-adr 1.3.12 → 1.3.13

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dzhechkov/skills-feature-adr",
-  "version": "1.3.12",
+  "version": "1.3.13",
   "description": "Adaptive Feature Development skill pack for Claude Code — 11-step pipeline with Complexity Router (S/M/L/XL), ADR-driven architecture, 15 agentic-qe skills, multi-agent fleet QE. Supports --full-qe, --full-qe-extended, --with-learning, and --knowledge-extractor modes.",
   "bin": {
     "skills-feature-adr": "./bin/cli.js"

package/templates/.claude/skills/feature-adr/SKILL.md

@@ -348,13 +348,19 @@ Step 0 (Complexity Router) checks:
 When `{AGENTIC_QE_MODE}` = `direct` | `direct-extended`, the pipeline runs a recall → store cycle
 over agentic-qe's MCP pattern memory (`namespace: "learning"`, `fleet_init` first):
 
-- **Step 0** — `memory_query("patterns/feature-adr/*")` → sets `{LEARNED_PATTERNS}` (top-3 by confidence); Step 1 folds them into the requirements brief as advisory lessons.
-- **Step 8** — recalls `patterns/feature-adr/qe/*` to prime the review checklist; stores a `qe-outcome` record after the gap loop closes.
-- **Step 9** — recalls `patterns/feature-adr/fleet/*` to prime agent focus; stores a `fleet-qe-finding` record after the fleet verdict.
-
-All calls are non-blocking — an error or empty result never stalls the pipeline. This layer is
+- **Step 0** — `memory_query("patterns/feature-adr/*")` **plus** `dz recall --json` over the durable
+  dz store → merged top-3 set `{LEARNED_PATTERNS}`; Step 1 folds them into the requirements brief as advisory lessons.
+- **Step 8** — recalls `patterns/feature-adr/qe/*` to prime the review checklist; after the gap loop
+  closes, stores a `qe-outcome` record in aqe **and** the same lesson via `dz teach` (durable).
+- **Step 9** — recalls `patterns/feature-adr/fleet/*` to prime agent focus; after the fleet verdict,
+  stores a `fleet-qe-finding` in aqe **and** the same lesson via `dz teach`.
+
+**Two stores by design:** the aqe kv store is *ephemeral across MCP-server restarts* (janitor/TTL —
+observed live), so it serves fast in-session semantic recall; the dz lexical store
+(`dz teach`/`dz recall`) is the durable half that survives to the next day. All calls are
+non-blocking — an error or empty result never stalls the pipeline. This layer is
 **distinct from the Keysarium reward layer** installed by `--with-learning`: it uses agentic-qe MCP
-memory, is gated on Direct modes only, and never touches `.keysarium/memory/`.
+memory + the dz store, is gated on Direct modes only, and never touches `.keysarium/memory/`.
 
 ### What changes with `--full-qe`
 

package/templates/.claude/skills/feature-adr/modules/00-complexity-router.md

@@ -86,8 +86,13 @@ Runs after tier classification, only when `{AGENTIC_QE_MODE}` = `direct` | `dire
 
 1. If the aqe fleet is not yet initialized this session, call `mcp__agentic-qe__fleet_init` first (required before any other aqe MCP tool).
 2. Call `mcp__agentic-qe__memory_query({ pattern: "patterns/feature-adr/*", namespace: "learning", limit: 5 })`.
-3. Select the top 3 results by `confidence` → set `{LEARNED_PATTERNS}`.
-4. Reference mode, no hits, or error → `{LEARNED_PATTERNS}` = `[]`; proceed normally.
+3. ALSO query the **durable half** of the loop — the dz lexical store:
+   `dz recall "<feature domain keywords>" --json --limit 5` (CLI; works wherever dz is set up).
+   Why both: the aqe kv store is **ephemeral across MCP-server restarts** (janitor/TTL — observed
+   live: a record survived hours, not days), so aqe serves in-session semantic recall while the
+   dz store is the durable memory the loop can trust tomorrow.
+4. Merge both result sets, dedupe by content, select the top 3 by confidence/reward → set `{LEARNED_PATTERNS}`.
+5. Reference mode, no hits, or errors (either store) → `{LEARNED_PATTERNS}` = `[]` or the surviving half; proceed normally.
 
 Step 1 folds `{LEARNED_PATTERNS}` into the requirements brief as "lessons from previous features" — advisory context only, never requirements themselves. Why here: Step 0 is the one point where recall can be keyed to the feature's phase and domain — per-prompt hook auto-injection cannot see which pipeline step is running.
 

package/templates/.claude/skills/feature-adr/modules/08-qe.md

@@ -226,6 +226,16 @@ mcp__agentic-qe__memory_store({
 
 Confidence must be honest — derived from the gap loop, never a fabricated 1.0
 (e.g. zero gaps in one iteration ≈ 0.9; gaps remaining after 3 iterations ≤ 0.4).
+
+Then persist the SAME lesson durably (the aqe kv store is ephemeral across MCP-server
+restarts — observed live; the dz lexical store is the half that survives):
+
+```
+dz teach "<the same one-sentence lesson>" --reward <same confidence> --domain feature-adr \
+  --type <success-pattern if verdict READY, else lesson-learned>
+```
+
+Non-blocking: if `dz` is not on PATH or the command fails, note it and proceed.
 Non-blocking: a write failure is logged and the pipeline continues.
 
 ## Output

package/templates/.claude/skills/feature-adr/modules/09-fleet-qe.md

@@ -294,6 +294,16 @@ mcp__agentic-qe__memory_store({
 })
 ```
 
+Then persist the SAME finding durably via the dz lexical store (the aqe kv store is
+ephemeral across MCP-server restarts — the dz half survives):
+
+```
+dz teach "<the same one-sentence lesson>" --reward <same confidence> --domain feature-adr \
+  --type <success-pattern if COMPLETE, else lesson-learned>
+```
+
+Non-blocking: if `dz` is unavailable or fails, note it and proceed.
+
 ## Direct Extended Mode (`--full-qe-extended`)
 
 When `{AGENTIC_QE_MODE}` = `direct-extended`, Step 9 gets everything from