Skip to content

feat: add parsimony plugin for token efficiency#12

Open
jm-cnaught wants to merge 2 commits intoCNaught-Inc:mainfrom
jm-cnaught:add-parsimony-plugin
Open

feat: add parsimony plugin for token efficiency#12
jm-cnaught wants to merge 2 commits intoCNaught-Inc:mainfrom
jm-cnaught:add-parsimony-plugin

Conversation

@jm-cnaught
Copy link

Summary

  • Adds parsimony plugin — a token efficiency skill that reduces controllable token usage by ~30% and speeds up responses by ~48% without sacrificing output quality
  • Includes skill (SKILL.md), two slash commands (/token-mode, /token-report), plugin manifest, and README with benchmark results
  • Registers parsimony in marketplace.json alongside the existing carbon plugin

What's included

File Purpose
plugins/parsimony/skills/parsimony/SKILL.md Core skill — tool rules, response rules, planning friction, token ledger
plugins/parsimony/commands/token-mode.md /token-mode <efficient|ultra|off> slash command
plugins/parsimony/commands/token-report.md /token-report savings summary slash command
plugins/parsimony/.claude-plugin/plugin.json Plugin manifest
plugins/parsimony/README.md Install instructions, modes, benchmarks
.claude-plugin/marketplace.json Updated to register parsimony
README.md Updated with parsimony in Plugins section

Benchmark results (Opus, controllable tokens only)

Task Savings Speed Gain
Summarize large file 80.9% 43% faster
Write TypeScript module 55.4% 58% faster
Debug Python function 34.6% 52% faster
Refactor JavaScript 19.3% 66% faster
Search/list files 17.1% 39% faster
Compare two files 4.4% 32% faster
Average 30.0% ~48% faster

Test plan

  • Install plugin via claude plugin install parsimony --marketplace cnaught-plugins
  • Verify skill loads and applies token-efficient behavior by default
  • Test /token-mode ultra and /token-mode off switching
  • Test /token-report output format
  • Verify no regressions in carbon plugin

🤖 Generated with Claude Code

Jacob Model and others added 2 commits February 26, 2026 07:29
Adds a new plugin that reduces controllable token usage by ~30% and
speeds up responses by ~48% without sacrificing output quality.

Includes:
- Parsimony skill with efficient/ultra/off modes
- /token-mode command for switching modes
- /token-report command for session savings summary

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…simony

- Strip rules already in Claude Code system prompt (parallel batching,
  use Glob/Grep directly, no preamble, code quality, YAGNI)
- Remove subagent model selection guidance (haiku/sonnet/opus) to
  explore separately as a feature
- Skill reduced from 63 to 47 lines

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
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