tldr

name

tldr

description

This skill should be used when the user asks for a TLDR, summary, recap, or condensed version of content. Invoked as "/tldr" with no arguments to summarize the previous Claude message from conversation context, or as "/tldr [topic]" to summarize whatever the user specifies. Handles requests like "summarize that", "give me the tldr", "recap what you just said", "shorten your last response", or "tldr of the error analysis".

tldr

Create a concise bullet-point summary of content from the current conversation.

Task

Mode 1: No arguments (/tldr)

Summarize the previous Claude assistant message.

1. Recall the previous assistant message from conversation context. You already have the full conversation history in your context window. Identify the last substantial assistant message — the one immediately before the user's /tldr invocation. Skip any messages that are only tool calls with no text output.

2. Validate:

   • If the last assistant message is very short (under 50 words), respond: "Last message is only N words — here it is in brief:" followed by a one-line summary.
   • If there is no previous assistant message (conversation just started), respond: "No previous message to summarize. Use /tldr [topic] to summarize specific content."

3. Analyze and extract key information (see guidelines below).

4. Format as bullets (see formatting below).

Mode 2: With arguments (/tldr [what to tldr])

Summarize whatever the user specifies.

1. Identify what to summarize. The user's argument describes or contains the content to summarize. This could be:

   • A topic discussed earlier: /tldr the error analysis
   • A reference to a specific message: /tldr your second response
   • Inline content the user pasted after the command

2. Locate the relevant content in the conversation context.

3. Analyze and extract key information (see guidelines below).

4. Format as bullets (see formatting below).

Extraction Guidelines

Focus on actionable and factual content:

• Critical findings — Root causes, discoveries, issues identified
• Actions recommended — Categorize by urgency (immediate/medium/long-term)
• Decisions made — Choices, options presented, recommendations
• Deliverables created — Files, documents, code written
• Next steps — User choices, follow-up actions required

Formatting Rules

• Maximum 8 bullet points total
• Each bullet: 1-2 sentences maximum
• Focus on "what" and "action" — omit explanations of "why"
• Use bold for category headers when grouping (optional)
• Prioritize: findings → actions → deliverables → next steps

Output Format

TLDR:

• [Key finding or root cause identified]
• [Action item with specific context]
• [Decision or recommendation made]
• [Deliverable: specific file or artifact created]
• [Next step or user choice required]

Examples

Example 1: Research Summary

**TLDR:**
- Root cause: Priority markers are advisory (level 4), not enforceable
- Claude 4.x prioritizes pragmatic goals over literal formatting rules
- Immediate action: Simplify buzzminson.md to <150 lines, add XML tags
- Medium action: Create PostToolUse hook for validation
- Deliverables: 3 research docs in docs/research/
- Next: Choose A) Templates, B) Hooks, C) Simplify, D) Review

Example 2: Code Implementation

**TLDR:**
- Fixed authentication bug in login flow (user.service.ts:45)
- Added JWT token refresh logic with 15-minute expiration
- Created middleware for route protection (auth.middleware.ts)
- Updated tests: 12 passing, 0 failing
- Next: Deploy to staging and test with production data

Example 3: With Arguments

User: /tldr the database migration discussion

**TLDR:**
- Migrating from PostgreSQL 14 to 16 requires pg_dump/pg_restore path
- Three tables need schema changes: users, sessions, audit_log
- Estimated downtime: 15 minutes with blue-green deployment
- Next: Run migration script on staging first

Error Handling

• No previous message: "No previous message to summarize. Use /tldr [topic] to summarize specific content."
• Very short message: Still produce a one-line summary instead of refusing.
• Ambiguous argument: "I found multiple discussions about [topic]. Which one? [list options]"
• Topic not found: "I don't see a discussion about [topic] in this conversation."

Implementation Notes

• Do NOT read files to access conversation history — use your context window
• Do NOT call external APIs — analyze using your own capabilities
• Do NOT execute external scripts
• Be concise — this command exists to save the user time