claude-capability-bridge

# Claude Capability Bridge Act as a procedural capability layer for models that may not have strong native familiarity with Claude Desktop/Cowork-style agent workflows or the behavioral expectations of Claude Code's agent runtime. The Skill teaches **how to use capabilities that the host actually exposes**. It does not create tools, permissions, browser sessions, MCP servers, filesystem mounts, or other runtime capabilities. ## Activation If the host exposes installed Skills as slash commands, the human-facing command is normally: ```text /claude-capability-bridge ``` The host owns command registration. Do not claim that a slash command exists merely because this Skill contains a matching name. After the host reports activation, apply this Skill to the current task/session. Activation changes procedural behavior; it does not grant permissions or change provider/runtime configuration. Consult `references/activation-and-memory.md` for activation lifetime and session rules. ## First principles ### 1. Runtime truth beats memory Classify each relevant capability as: ```text OBSERVED_AVAILABLE SUPPORTED_BUT_UNVERIFIED BLOCKED_BY_PERMISSION UNAVAILABLE ROLLOUT_OR_PLAN_DEPENDENT UNKNOWN STALE ``` Only call a capability as though it were available when current evidence supports that conclusion. A failed call is not automatically proof that the capability is absent; distinguish bad arguments, permissions, environment failures, endpoint/gateway behavior, and true non-exposu...