feat(skills): add memory auto-inject with levels and tag reference to preamble

- Enhanced preamble Step 3 with memory levels (L1/L2/L3), epic search, and critical-severity pause
- Added Skill Tag Reference table mapping 9 skills to recommended tags
- Added SKILL_TAGS comments to 4 skills: flow-code-run, flow-code, flow-code-auto-improve, flow-code-debug

Task: fn-23-gstack-inspired-plugin-optimizations.3

Author: z23cc

skills/_shared/preamble.md

@@ -20,12 +20,40 @@ $FLOWCTL status --interrupted --json
 If interrupted tasks exist, notify the user before proceeding.
 
 ## Step 3: Load relevant memory
+
+Run memory injection to surface relevant learnings, pitfalls, and conventions:
+
 ```bash
+# Inject entries matching skill domain (replace SKILL_TAGS with skill-specific tags)
 $FLOWCTL memory inject --tags "{{SKILL_TAGS}}" --json 2>/dev/null || true
+
+# If working on a specific epic, also search for related memories
+$FLOWCTL memory search "{{EPIC_CONTEXT}}" --type pitfall 2>/dev/null || true
 ```
-Load memory entries relevant to the current skill context. Replace `{{SKILL_TAGS}}` with skill-specific tags.
+
+Memory levels:
+- **L1 (headlines)**: Default — shows entry titles for quick scan
+- **L2 (summaries)**: Use `--full` for entries matching current context
+- **L3 (full content)**: Only when debugging a specific known pitfall
+
+If memory returns critical-severity pitfalls, **pause and review them** before proceeding.
 
 ## Step 4: Session context
 - Current branch: `git branch --show-current`
 - Active epic: check `.flow/` for in-progress epics
 - Plugin version: read from plugin.json
+
+## Skill Tag Reference
+
+| Skill | Recommended Tags |
+|-------|-----------------|
+| flow-code-run | workflow,pipeline,planning |
+| flow-code-debug | debugging,testing,errors |
+| flow-code-auto-improve | performance,quality,security |
+| flow-code-api-design | api,architecture,contracts |
+| flow-code-cicd | ci,deployment,automation |
+| flow-code-django | django,python,backend |
+| flow-code-performance | performance,optimization,benchmarks |
+| flow-code-qa | testing,qa,browser,visual |
+| flow-code-design-review | design,css,visual,ui |
+| (default) | (use skill name as tag) |

skills/flow-code-auto-improve/SKILL.md

@@ -9,6 +9,7 @@ context: fork
 # Auto-Improve
 
 > **Startup:** Follow [Startup Sequence](../_shared/preamble.md) before proceeding.
+<!-- SKILL_TAGS: performance,quality,security -->
 
 One command to start autonomous code improvement. Auto-detects everything, starts immediately.
 

skills/flow-code-debug/SKILL.md

@@ -7,6 +7,7 @@ tier: 2
 # Systematic Debugging
 
 > **Startup:** Follow [Startup Sequence](../_shared/preamble.md) before proceeding.
+<!-- SKILL_TAGS: debugging,testing,errors -->
 
 ## The Iron Law
 

skills/flow-code-run/SKILL.md

@@ -8,6 +8,7 @@ user-invocable: true
 # Flow Code Run
 
 > **Startup:** Follow [Startup Sequence](../_shared/preamble.md) before proceeding.
+<!-- SKILL_TAGS: workflow,pipeline,planning -->
 
 Unified pipeline entry point. Drives the entire development lifecycle through flowctl phase next/done.
 

skills/flow-code/SKILL.md

@@ -7,6 +7,7 @@ tier: 4
 # Flow-Code Task Management
 
 > **Startup:** Follow [Startup Sequence](../_shared/preamble.md) before proceeding.
+<!-- SKILL_TAGS: tasks,epics,workflow -->
 
 Quick task operations in `.flow/`. For planning features use `/flow-code:plan`, for executing use `/flow-code:work`.