What problem does it solve? Evaluating a plugin requires real evidence, but raw run logs can be misleading — a number without its denominator can be wrong by orders of magnitude. This Skill computes the trace evidence block and conformance results for a plugin version deterministically, with an explicit sufficiency verdict, so no model ever guesses at the data. ## Core Features & Use Cases - Trace Evidence Block: Counts usable runs, returns, unplaced steps, per-tool calls and refusals, and never-called tools, all derived from the event log with explicit coverage denominators. - Sufficiency Verdict: Reports whether the evidence is sufficient for judging (5 usable runs floor), while treating a thin trace block as a valid outcome rather than a failure. - Three-Valued Conformance: Reports R1–R14 as true, false, or not_measured, refusing to score clauses the contract does not measure. - Use Case: After locating a third-party plugin with no local run history, run plugin_conform to get a starting ruler of which requirements are measured, then decide whether there is enough evidence to judge. ## Quick Start Run the profile stage for the located plugin and version to compute its trace evidence block, coverage, and conformance verdict.
