The kit targets Claude Code only

Context and Problem Statement

The kit is built out of mechanisms only one harness provides: a PreToolUse hook manifest, a plugin .mcp.json that ships its own MCP servers, slash commands under commands/, and skills that resolve each other through _shared/ relative links. Portability to other agent harnesses was evaluated in the v2 meta review and declined. (This decision previously lived nowhere — it was implied by the directory layout, which is why it is recorded here.)

Considered Options

• Target Claude Code only, and use its mechanisms without abstraction.
• Abstract the harness behind an adapter so the skills could run elsewhere.
• Ship the skills as harness-neutral prose and leave the wiring to the consumer.

Decision Outcome

The kit targets Claude Code and only Claude Code: hooks, the bundled MCP servers, slash commands and inter-skill relative links are used directly, with no portability layer, because every abstraction that would make them harness-neutral would also make them weaker in the one harness that is actually supported.

Consequences

Another harness cannot run the kit without reimplementing its wiring, and a Claude Code mechanism that changes shape is a breaking change for the whole kit rather than for one adapter. In exchange the skills can rely on hooks, sub-agents and MCP servers being present rather than probing for them.