Skip to content

feat(caveman): extend write-normal carve-out to issue/PR/MR titles and descriptions#483

Open
paolomainardi wants to merge 1 commit into
JuliusBrussee:mainfrom
paolomainardi:feat/issue-pr-mr-prose-carveout
Open

feat(caveman): extend write-normal carve-out to issue/PR/MR titles and descriptions#483
paolomainardi wants to merge 1 commit into
JuliusBrussee:mainfrom
paolomainardi:feat/issue-pr-mr-prose-carveout

Conversation

@paolomainardi
Copy link
Copy Markdown

πŸ€– This was written by an AI agent on behalf of @paolomainardi.

Summary

The active-mode reminder and skill boundaries already tell the model to write code, commits, and security warnings in normal prose. They say nothing about issue, pull-request, and merge-request titles and descriptions β€” which are prose artifacts that should not be compressed into caveman fragments. In practice the model writes terse fragmented PR/MR/issue text while caveman mode is active.

This extends the existing boundary line to explicitly cover issue/PR/MR titles and descriptions, across every source-of-truth surface that carries the carve-out (CI mirrors under plugins/ are left untouched and regenerate from these).

Changes

  • src/hooks/caveman-mode-tracker.js β€” per-turn UserPromptSubmit reinforcement reminder
  • src/hooks/caveman-activate.js β€” SessionStart full ruleset ## Boundaries
  • skills/caveman/SKILL.md β€” caveman skill behavior doc ## Boundaries
  • src/rules/caveman-activate.md β€” auto-activation rule body (Cursor/Windsurf/Cline/Copilot)
  • src/tools/caveman-init.js β€” per-repo init template
  • src/plugins/opencode/plugin.js and src/plugins/opencode/commands/caveman.md β€” opencode reminder + command

Each edit changes the single boundary line, e.g. Code/commits/security: write normal. becomes Code/commits/security/issue+PR+MR titles+descriptions: write normal prose.

Testing

  • node --test tests/installer/*.test.mjs β€” 51 pass, 0 fail
  • node --check on all edited JS hooks β€” clean

…d descriptions

The active-mode reminder and skill boundaries told the model to write code, commits, and security warnings in normal prose, but said nothing about issue, pull-request, and merge-request titles and descriptions. Those are prose artifacts that should not be compressed into caveman fragments.

Extend the boundary line across all source-of-truth surfaces: the per-turn UserPromptSubmit reminder (src/hooks/caveman-mode-tracker.js), the SessionStart ruleset (src/hooks/caveman-activate.js), the caveman skill doc (skills/caveman/SKILL.md), the auto-activation rule body (src/rules/caveman-activate.md), the per-repo init template (src/tools/caveman-init.js), and the opencode plugin reminder and command (src/plugins/opencode/).

Assisted-by: claude-code/claude-opus-4-8
@paolomainardi
Copy link
Copy Markdown
Author

πŸ€– This was written by an AI agent on behalf of @paolomainardi.

Tracking issue: #484

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant