Skip to main content
Generalmajesticlabs-dev

reflect

Reflect on the current session and suggest AGENTS.md improvements based on user corrections, repeated feedback, and patterns observed during interaction. Use when ending a session to encode lessons learned. Don't use for cross-session learning from git history - use `learn` instead.

Stars
39
Source
majesticlabs-dev/majestic-marketplace
Updated
2026-05-13
Slug
majesticlabs-dev--majestic-marketplace--reflect
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/majesticlabs-dev/majestic-marketplace/HEAD/plugins/majestic-tools/skills/reflect/SKILL.md -o .claude/skills/reflect.md

Drops the SKILL.md into .claude/skills/reflect.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

Reflect

Analyze the current session and suggest improvements to AGENTS.md based on patterns, feedback, and lessons learned.

For cross-session learning from git history and handoffs, use Skill("learn") instead.

Data Sources

Source What to Extract
Conversation history User corrections, repeated feedback, preferences
.agents/session_ledger.md Key Decisions, patterns from current work (if exists)

Process

Step 1: Review Conversation History

Analyze all messages in this conversation:

  • User feedback and corrections
  • Moments where instructions weren't followed
  • Repeated issues or preferences
  • Patterns that emerged during interaction

Step 2: Check Session Ledger (if exists)

[ -f ".agents/session_ledger.md" ] && cat ".agents/session_ledger.md"

Extract any Key Decisions or patterns documented there.

Step 3: Identify Improvement Opportunities

Look for:

  • Gaps in current AGENTS.md that led to misunderstandings
  • User preferences that should be encoded
  • Patterns that could prevent future issues
  • Corrections that apply broadly (not one-off incidents)

Step 4: Generate Specific Improvements

Create actionable, specific edits. Each suggestion should:

  • Be concrete (not vague principles)
  • Prevent similar issues in future
  • Address patterns, not one-off incidents

Step 5: Present Suggestions

Use this format:

## Session Reflection

### 1. [Pattern Name]
**Gap**: What's missing or unclear in AGENTS.md
**Proposed**: Specific text to add
**Reasoning**: Evidence from this session (e.g., "You corrected this 3 times")

### 2. [Pattern Name]
...

Step 6: Apply Changes (if approved)

Use AskUserQuestion:

  • "Would you like me to apply these improvements to AGENTS.md?"

If yes:

  • Use Edit tool to update AGENTS.md
  • Confirm changes were applied

Guidelines

  • Only suggest improvements that genuinely enhance behavior
  • Keep suggestions specific and actionable
  • Don't suggest redundant rules
  • Focus on patterns, not one-off incidents
  • Maintain existing AGENTS.md structure and style

Example Output

## Session Reflection

### 1. Error Handling Preference
**Gap**: No guidance on error handling style
**Proposed**: "Always use try-catch blocks with specific error types rather than generic catches"
**Reasoning**: You corrected this pattern 3 times during our session

### 2. Import Organization
**Gap**: No import organization rules
**Proposed**: "Group imports by: 1) External packages, 2) Internal modules, 3) Types"
**Reasoning**: You consistently reorganized imports this way

Would you like me to apply these improvements to AGENTS.md?