A token efficient personality runtime built for DeepSeek Harness
Freakz2z/dsh-catgirl-plugin · skills
dsh-catgirl-plugin A token-efficient personality runtime built for DeepSeek Harness.
★ 5GitHub stars
0Forks
2026-09-06Last updated
JavaScriptLanguage
MITLicense
Key features
- The traditional approach is to let LLM "play the cat girl" by itself: a long persona worth hundreds of tokens is injected into the system prompt, each request is repeatedly paid, and the model output becomes verbose.
- Degenerate BoundaryTasks that require specialized tools (subagent parallelism, skill calls) - the model changes the policy.
- Reasoning Router needs to establish a cross-model quality baseline first and is not automatically enabled by default.
- This plug-in does the opposite - Persona Virtualization
Requirements
- The headless one-time process does not wait for the background subagent: the subagent with run_in_background: true is terminated when the parent process exits (headless application limitation, Web UI does not have this problem)
Install command
dsh plugin --profile web add github:Freakz2z/dsh-catgirl-plugin