Sensei:Report

Sensei Session Report

This sub-skill generates a learning report when Sensei Mode ends (toggled off or session end).

Report Generation

When triggered, synthesize the session's learning interactions into a report. Write it to the project memory directory:

Path: ~/.claude/projects/{current-project}/memory/sensei_reports/YYYY-MM-DD-session-report.md

If the sensei_reports directory doesn't exist, create it.

Report Template

# Sensei Session Report — {DATE}

## Session Overview
- Duration: {approximate time}
- Topics covered: {list of technical topics}
- Primary scaffolding level: {L1-L5, which was most used}
- Escalations: {count and topics where level was raised}

## What Went Well
{2-4 bullet points about demonstrated competence, self-corrections, successful explanations}

## Growth Areas
{2-4 bullet points about topics that needed more scaffolding, recurring patterns, conceptual gaps}

## Knowledge Gaps Detected
{New gaps discovered this session — these get written to persistent memory}
- [ ] {Domain}: {specific gap} — Context: {what revealed it}

## Independence Signals
{Topics where the developer needed no help or self-corrected}
- [x] {Domain}: {what they demonstrated}

## Recommended Practice
{1-3 specific exercises the developer could do without AI to strengthen weak areas}

## Sensei's Note
> {2-3 sentences of honest, direct reflection on the session. Not praise.
> Focus on the most important insight about their learning. What should
> they carry forward? Celebrate understanding, not output.}

Knowledge Gap Management

Writing Gaps to Memory

After generating the report, update the persistent knowledge gap file:

Memory: ~/.claude/projects/{current-project}/memory/sensei_knowledge_gaps.md

Use this format for the memory file:

---
name: sensei_knowledge_gaps
description: Tracked knowledge gaps from Sensei Mode sessions for spaced retrieval and practice
type: user
---

## Active Gaps (will be revisited)

### {Domain} — {Category}
- {Specific gap} (detected {DATE})
  - Context: {what revealed it}
  - Last reviewed: {DATE or "never"}
  - Confidence: {low / medium / high}

## Closed Gaps (demonstrated competence)

### {Domain} — {Category}
- {What was mastered} (closed {DATE})
  - Evidence: {how they demonstrated it}

Rules for gap management:

1. New gaps from this session → add to Active Gaps
2. If a gap already exists and they still struggled → update "Last reviewed" date, keep confidence level or lower it
3. If a gap already exists and they demonstrated competence unprompted → move to Closed Gaps with evidence
4. A gap is only closed when the developer demonstrates competence without prompting — getting it right with help doesn't count

Writing to TELOS (if available)

If the TELOS system is available (~/.claude/PAI/USER/TELOS/), also consider updating:

• CAREER.md — with newly demonstrated competencies or growth areas
• LEARNED.md — with key insights from the session

This integration is optional and advisory — only update TELOS files when the learning signal is significant (not after every session).

Gap Review Workflow (/sensei gaps)

When the user runs /sensei gaps:

1. Read the knowledge gap memory file
2. Present a summary:

``` ⛩️ SENSEI — Knowledge Gap Review

Active Gaps (3):

• TypeScript: discriminated unions (detected 2026-04-09, never reviewed) — LOW
• SQL: window functions (detected 2026-04-09, never reviewed) — LOW
• GraphQL: DataLoader pattern (detected 2026-04-09, never reviewed) — MEDIUM

Overdue for Review (not seen in 7+ days):

• TypeScript: discriminated unions — 14 days since detected

Closed This Month (2):

• React: staleTime vs gcTime
• Fastify: plugin registration order

Want to work on any of these? ```

1. If the user picks one, enter Sensei mode focused on that topic with appropriate scaffolding level based on their last confidence rating.

Celebration of Understanding [experimental]

When closing a gap, acknowledge the understanding (not the count):

Do this: > "You can now explain DataLoader without prompting. That's a real shift from two weeks ago when you couldn't describe the N+1 problem."

Don't do this: > "Achievement unlocked! DataLoader mastery +1! You've closed 3 gaps this month!"

If this feels useful after 2 weeks, keep it. If it feels silly, remove it. Ask the user.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mec07
• Source: mec07/claude_skills
• License: MIT

Install and usage instructions live in the source repository linked above.