fix: caveman full/ultra levels now use complete sentences so host system prompts (Claude Code) stop overriding the style - #788
Open
ly6751 wants to merge 1 commit into
Conversation
…tem prompts (Claude Code) stop overriding the style Root cause (issue JuliusBrussee#687, analysis in issue comment): Claude Code's system prompt explicitly tells the model to 'write in complete sentences' and not to 'compress the writing into fragments, abbreviations, arrow chains or jargon'. That instruction lives in the system prompt and outranks hook-delivered context, so caveman's fragment-based full/ultra levels lost every conflict and mode had no visible effect. Fix: compression now comes from vocabulary and selectivity (dropping filler, pleasantries, hedging, recaps, restated tool output, options not chosen) - never from fragments or dropped articles. Applied across all surfaces: - skills/caveman/SKILL.md + plugins/caveman/skills/caveman/SKILL.md (synced copy) - src/rules/caveman-activate.md + RULE_BODY fallback in src/tools/caveman-init.js - src/hooks/caveman-activate.js + src/hooks/caveman-mode-tracker.js (Claude Code) - src/plugins/opencode/plugin.js + opencode command - commands/caveman.md, skills/caveman-help/SKILL.md, skills/caveman/README.md - Added Model compatibility section documenting the host-prompt conflict Verified: tests/test_hooks.py (7), tests/test_mode_tracker.py (23), tests/test_caveman_init.js (1), tests/test_mode_tracker_stdin.js (2) all pass.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #687
Root cause: Claude Code's system prompt explicitly instructs the model to write in complete sentences and not compress writing into fragments/abbreviations. That instruction lives in the system prompt and outranks hook-delivered context, so caveman's fragment-based full/ultra levels lost every conflict — mode had no visible effect.
Fix: Compression now comes from vocabulary and selectivity (dropping filler, pleasantries, hedging, recaps, restated tool output, options not chosen) — never from fragments or dropped articles. Applied across all surfaces:
Verified: tests/test_hooks.py (7), tests/test_mode_tracker.py (23), tests/test_caveman_init.js (1), tests/test_mode_tracker_stdin.js (2) all pass.