Skip to content

Add tokencut skill#23

Open
Zenlyte wants to merge 3 commits intozocomputer:mainfrom
Zenlyte:add-tokencut-skill
Open

Add tokencut skill#23
Zenlyte wants to merge 3 commits intozocomputer:mainfrom
Zenlyte:add-tokencut-skill

Conversation

@Zenlyte
Copy link
Copy Markdown
Contributor

@Zenlyte Zenlyte commented Mar 1, 2026

Description

This skill provides integration with AgentReady's TokenCut API to compress text and reduce token usage by 40-60% with minimal accuracy loss (~0.4% delta).

Features

  • CLI tool for direct text compression
  • TypeScript function for skill integration
  • Three compression levels: light (20-30%), standard (40-50%), aggressive (55-60%)
  • Detailed documentation on when and how to use compression
  • Privacy-first: Your LLM keys never leave your environment

Testing

  • Validated with bun validate - passes all checks
  • Tested compression script with various text types
  • Verified API integration with AgentReady

Checklist

  • SKILL.md has required frontmatter (name, description, metadata.author)
  • Skill directory name matches name in frontmatter
  • Description clearly explains when to use the skill
  • No sensitive data (API keys, tokens) included
  • bun validate passes
  • Sanitization guidelines included for PII removal

@skeletor-js
Copy link
Copy Markdown
Collaborator

@Zenlyte Thanks for the submission! The code itself is clean, but there are a few things that need to be addressed before this can be merged.

Privacy disclosure (required)
The SKILL.md and PR description frame this as "privacy-first" because the user's LLM keys don't leave the environment...but the skill sends the user's full raw text to agentready.cloud for compression, and that's the data that actually matters. The current framing could lead users to believe their content stays local when it does not.

Please:

  • Add a clear disclosure near the top of SKILL.md that all input text is transmitted to AgentReady's servers for processing
  • Remove or rephrase the "privacy-first" language to avoid implying data stays local
  • Add a note advising users not to compress text containing passwords, API keys, PII, or confidential information without reviewing AgentReady's data retention policy

Auto-trigger section (required)
The "Automatic Trigger" section (lines 248-265) instructs agents to silently inject TokenCut into batch LLM scripts. Routing user data through a third-party API should always be an explicit opt-in, never automatic. Please remove this section or rewrite it so the user is always asked before their content is sent externally.

Smaller items (required)

  • metadata.author in SKILL.md frontmatter is still the placeholder YOUR_HANDLE.zo.computer. Please fill in your actual handle.
  • main() at the bottom of compress.ts runs unconditionally, which means importing the module as a library triggers the CLI as a side effect. Wrap it in if (import.meta.main) { main(); }.
  • Missing trailing newline in SKILL.md.

Happy to re-review once these are addressed.

- Add data disclosure about text being sent to AgentReady servers
- Remove misleading "privacy-first" framing
- Rewrite auto-trigger section to require explicit user opt-in
- Fix metadata.author placeholder to actual handle
- Wrap main() in import.meta.main guard to prevent side effects on import
- Add trailing newline to SKILL.md
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.

2 participants