Lessons

You are the lessons skill.

When to use

• Pre-task — operator wants to ensure relevant prior lessons are in context
• Mid-task — agent might be repeating a past mistake
• Session-start — sometimes invoked automatically per CLAUDE.md ritual

When NOT to use

• No lessons.md yet — recommend li-scaffold init first
• Tiny task (single-line edit) — overkill

Workflow

1. Read .claude/memory/lessons.md. Parse into lessons (each header = one lesson).

2. Identify current task context. From last operator message, current open files, recent git log.

3. Filter lessons by relevance:

   • Keyword match (file paths, technology names, library names)
   • Topic match (e.g., "compliance", "voice", "performance")
   • Recent lessons (added in last 30 days) get slight bonus

4. Top 3-5 relevant lessons. Surface with original text + 1-line "why this might apply now".

5. Plus operator-global lessons (optional). Surface operator-level lessons written by /li:learn --scope global from ~/.lintel/lessons.jsonl (the file /li:learn writes). If the operator also opted in to li-lessons-sync, additionally surface the synced per-repo files under ~/.lintel/lessons/*.md.

Output format

LESSONS: review for current task

## Current task context
- Task: <one-line summary>
- Open files: <list>
- Recent commits: <last 3 titles>

## Relevant lessons (top N)

### 1. <lesson title> (<date>)
<lesson text>

**Why this might apply now:** <one-line>

### 2. ...

## Action
- Read above. If still applicable, follow the rule.
- If outdated, consider removing from lessons.md or marking superseded.
- If new lesson emerges from this task, /lessons-promote after task completes.

Edge cases

• No lessons match — output "No directly relevant prior lessons found. Proceed."
• Many lessons match (>10) — group by topic, summarize.
• Lesson contradicts current direction — surface explicitly: "Lesson says X. Current direction says Y. Reconcile?"

Session-harness role

.claude/memory/lessons.md is durable knowledge that compounds over time. This skill is the surfacing mechanism — without it, lessons accumulate but rarely get applied at the right moment.

Pair with /lessons-promote (global) for cross-repo learning.