Dsh Continual Harness
jasen215/dsh-continual-harness · productivity
DeepSeek Harness (DSH) plugin for self-improving AI agents: continual learning, persistent memory, cross-session knowledge, review-and-refine workflows, and automatic rollback.
★ 7GitHub stars
1Forks
2026-09-04Last updated
TypeScriptLanguage
MITLicense
Key features
- State projection (inject harness context each step): agent/pre-step waterfall listener
- Skills are real dsh skillsapplied skill edits materialize as /SKILL.md bundles (with provenance metadata) under Config.skillsDir, kept in sync by deletes/rollbacks without touching user-owned skills in the same directory.
- harness_refine — mode'plan' (default) plans from instructions and commits atomically
- incremental injection when the content digest changes
- Review and automatic refinementsession/event listener on turn interval / compaction end
Requirements
- peer floors stay =0.1.0-rc.6, so older dsh releases keep working.
- Since dsh 0.1.2-alpha.3 no longer provides @deepseek-ai/dsh-home-paths inside the profile bundle, the plugin declares it as a hard dependency
Install command
dsh plugin --profile web add github:jasen215/dsh-continual-harness