botharness / dsh-plugin-dev
Establish precise DeepSeek Harness (DSH) and Cordis vocabulary and choose stable architectural seams. Use when designing or reviewing DSH plugins, distinguishing native from application-defined concepts, or deciding among Service, Event, Registry, SessionEvent, Projection, persistence, and execution boundaries.
DSH/Cordis context and decisions
Skill v0.3.4 · verified against DSH 0.1.6-alpha.2 (upstream SHA ddefc45fbc7f8e46dd73185e68295696d1297887).
Foundation-first workflow
- Read [`references/context.md`](references/context.md) completely. Use its canonical leading words so every important object and boundary maps to one defined DSH/Cordis term. Label downstream concepts as application-defined.
- Follow [`references/decision-tree.md`](references/decision-tree.md). Give each responsibility one primary seam with explicit ownership and lifecycle.
- After choosing the seam, verify version-specific API names, signatures, manifests, build behavior, and runtime details against the current DSH official documentation, pinned upstream source, and the running Host.
Scope boundary
This skill intentionally contains only the stable semantic and architectural layer: Context, canonical vocabulary, concept maps, and decision logic. It does not cache a Host API catalog, Client API catalog, Slot inventory, or community implementation survey. Those details change too quickly to be a trustworthy installed skill.
Historical investigations remain in the canonical BotHarness repository under docs/research/; they are evidence for maintainers, not published guidance. A downstream product's Context, architecture, and ADRs remain outside this skill and own its product vocabulary.
References
references/context.md— canonical DSH/Cordis vocabulary, boundaries, and conceptual architecture.references/decision-tree.md— requirement-to-seam decisions for dispatch, persistence, execution, and presentation.