| name | session-analyzer |
| description | Analyzes a session summary and extracts reusable patterns as proposed SKILL.md drafts. Trigger: /analyze-session, or user says "extract patterns", "make this a skill", "what can we reuse from this session".
|
| metadata | {"author":"javi-ai","version":"1.0","tags":["learning","skills","patterns","meta"],"category":"workflow"} |
| allowed-tools | Read, Write, mcp__plugin_engram_engram__mem_search, mcp__plugin_engram_engram__mem_get_observation |
Purpose
After a productive session, extract repeatable patterns and propose them as new SKILL.md files.
The user reviews and approves; approved skills are saved to ~/.claude/skills/ via javi-ai propose approve.
Extraction Algorithm
Step 1: Load session context
If triggered post-session, load the session summary from Engram:
mem_search(query: "session summary", project: "{project}")
If triggered manually, ask the user: "Share the session summary or describe what pattern you want to capture."
Step 2: Identify patterns
Look for:
- Repeated solutions — same approach used 3+ times in the session
- Non-obvious conventions — decisions that required research or deliberation
- Tool combinations — specific sequences of tools that solved a class of problem
- Anti-patterns avoided — things explicitly NOT done and why
Step 3: For each pattern found, draft a SKILL.md
Use this structure:
---
name: {kebab-case-name}
description: >
{one-line description of when to use this skill}
Trigger: {trigger phrases}
metadata:
author: javi-ai
version: "0.1-proposed"
tags: [{relevant tags}]
category: {workflow|testing|architecture|debugging}
allowed-tools: {tools the skill needs}
---
## Purpose
{why this skill exists — the problem it solves}
## Steps
{numbered steps}
## Rules
1. {critical constraint}
2. {critical constraint}
Step 4: Propose via javi-ai
Save each draft to ~/.javi-ai/proposed/{skill-name}.md and report:
Extracted {N} patterns from session:
1. {skill-name} — {one-line description}
→ Save proposal: javi-ai propose approve {skill-name}
→ Discard: javi-ai propose reject {skill-name}
2. ...
Rules
- Quality over quantity — propose 1-3 high-quality skills, not 10 mediocre ones
- Specificity wins — "how to configure Biome for Ink projects" beats "linting"
- Include trigger phrases — a skill with no trigger is never loaded
- Version as 0.1-proposed — marks it as unreviewed
- Don't propose duplicates — check existing skills first with mem_search
