@champpaba/claude-agent-kit 3.0.2 → 3.0.3

package/.claude/agents/_shared/pre-work-checklist.md

@@ -131,6 +131,52 @@ Reporting to Main Claude...
 
 ---
 
+### Step 0.6: Memory Context Query (v2.2.0 - claude-mem Integration)
+
+**Before implementation, query claude-mem for related past work:**
+
+WHY: Avoid repeating mistakes, leverage past decisions, maintain consistency.
+
+1. **Identify current task keywords:**
+   - Component name (e.g., "authentication", "payment")
+   - Technology (e.g., "JWT", "Stripe")
+   - Pattern type (e.g., "form validation", "API endpoint")
+
+2. **Query past observations:**
+   Ask naturally - mem-search skill auto-invokes:
+   ```
+   "What decisions were made about {component}?"
+   "Did we solve similar problems with {technology}?"
+   "Any past bugs related to {pattern}?"
+   ```
+
+3. **Apply relevant learnings:**
+   - If past decision exists → Follow established pattern
+   - If past bug was fixed → Apply the fix
+   - If conflict with current spec → Note and report to Main Claude
+
+4. **Report in validation:**
+   ```markdown
+   Memory Context: (Step 0.6)
+   - [ ] Queried claude-mem for related past work
+   - Relevant observations found:
+     - [#ID] {summary} → Will apply: {how}
+   - No relevant observations: Proceeding fresh
+   ```
+
+**Example:**
+```
+Task: Implement refresh token endpoint
+
+Query: "decisions about refresh tokens"
+Result: #12345 - "Chose rotating refresh tokens with Redis storage"
+
+→ Apply: Use rotating tokens + Redis (not in-memory)
+→ Report: "Following decision #12345 - rotating tokens"
+```
+
+---
+
 ### Step 1-4: Standard Checks
 
 1. **Context Discovery** - Load project context via agent-discovery.md
@@ -159,6 +205,12 @@ Design Spec Implementation: (Step 5)
   - {requirement 1 from design.md}
   - {requirement 2 from design.md}
 
+Memory Context: (Step 0.6 - claude-mem)
+- [ ] Queried claude-mem for related past work
+- Relevant observations:
+  - {#ID: summary → will apply how}
+  - OR: No relevant observations found
+
 Project Context:
 - Project: {name}
 - Stack: {tech-stack}
@@ -178,6 +230,7 @@ Ready to Implement:
 - [ ] Task understood
 - [ ] Dependencies identified
 - [ ] Required libraries identified
+- [ ] Memory context applied (if available)
 - [ ] Approach planned (using specified libraries)
 ```
 

package/.claude/commands/csetup.md

@@ -158,6 +158,71 @@ if (fileExists(projectStatusPath)) {
 }
 ```
 
+### Step 1.6: Memory Context Query (v2.2.0 - claude-mem Integration)
+
+**WHY:** Query past work to leverage decisions, avoid repeating mistakes, and maintain consistency.
+
+```typescript
+// Extract keywords from change-id and proposal title
+const changeKeywords = changeId.split('-').join(' ')
+const proposalPath = `openspec/changes/${changeId}/proposal.md`
+const proposalContent = fileExists(proposalPath) ? Read(proposalPath) : ''
+const proposalTitle = proposalContent.match(/^#\s+(.+)/m)?.[1] || changeId
+
+output(`\n🧠 Querying claude-mem for related past work...`)
+
+// Use mem-search skill to find related observations
+// The skill auto-invokes when asking about past work
+const queries = [
+  `decisions about ${changeKeywords}`,
+  `bugs related to ${changeKeywords}`,
+  `implementations of ${changeKeywords}`
+]
+
+// Claude will auto-invoke mem-search for these queries
+// Results are stored for inclusion in research-checklist.md
+
+let pastLearnings = []
+
+// Note: In practice, Main Claude asks these questions naturally
+// and mem-search skill returns relevant observations
+
+output(`   Searched for: ${changeKeywords}`)
+output(`   (Results will be included in research-checklist.md if relevant)`)
+output(``)
+
+// Store for later use in research-checklist.md generation
+// pastLearnings will be populated by mem-search results
+```
+
+**Integration with research-checklist.md:**
+
+When generating `research-checklist.md` (Step 2.6), include a "Past Learnings" section:
+
+```markdown
+## Past Learnings (from claude-mem)
+
+> Related observations from previous sessions:
+
+| ID | Type | Summary | Relevance |
+|----|------|---------|-----------|
+| #12345 | decision | Chose Drizzle over Prisma | HIGH |
+| #12340 | bugfix | Fixed N+1 query in user list | MEDIUM |
+
+### Key Takeaways:
+- Use Drizzle patterns established in #12345
+- Watch for N+1 queries (see #12340 for solution)
+```
+
+If no relevant observations found:
+```markdown
+## Past Learnings (from claude-mem)
+
+No related past work found. Proceeding fresh.
+```
+
+---
+
 ### Step 2: Read OpenSpec Files
 
 Read in order: