Menu
Think-before-writing rule for content-generating skills. Complete all reasoning before producing any file content. Auto-consulted before any skill writes file content.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Add a dependency relationship between GitHub issues using the native addBlockedBy mutation
Reusable file sequence numbering with prefix (NNNN-name) and suffix (name-NNNN) modes. Returns the next available zero-padded sequence number for a given directory.
Drop a timestamped note for the current issue - collected by /finish-issue for PR descriptions
Create or update CHANGELOG entries with tone guardrails, thematic grouping, and implementation-detail leak detection.
Delete an issue's working directory (.claude-work/issues/<ID>/) after confirming with the user via interactive prompt
Create a commit message file in .claude-work/commit-msgs/ with auto-numbered filenames. Focuses on WHY not WHAT. The diff already shows what changed. User reviews and commits manually.
| name | pre-write |
| version | 2026.07.27@6dc7d1d |
| description | Think-before-writing rule for content-generating skills. Complete all reasoning before producing any file content. Auto-consulted before any skill writes file content. |
| user-invocable | false |
| allowed-tools | null |
Consulted automatically by content-generating skills. Not invoked directly by the user.
Complete all context-gathering and reasoning before producing any file content. The file must reflect finished thinking, not in-progress deliberation.
Phrases like "oh wait," "I now realize," "actually," "hmm," or any mid-stream self-correction in a generated file are signals that writing started too early. They must not appear in generated files.
If context is incomplete — or if a decision would benefit from user input — stop and use /question before proceeding. Never embed uncertainty or evolving reasoning in the generated file.
The test: can you state the full shape of the output before writing the first word? If not, gather more context first.
