analyse-governance-statslisted

# Analyse Governance Statistics Read this session's own governance spine and explain it. Not a dashboard tour — an analysis: what was called, what was permitted, what was refused, what it cost, and what that pattern says about how the agent behaved. ## When to Use Use this after some activity has happened in the session — a few skills run, a few tools called, ideally at least one denial. Use it when someone asks "how do you know what the agent did?" or "what does this actually cost?". Running it against an empty session is a fine demonstration of an empty table and nothing else. Do something first. ## How to Use ### 1. Pull the numbers ``` mcp__systemprompt__governance_stats {} ``` The result covers only the calling identity's own activity. There is no deployment-wide view here and that is deliberate: authority in this terminal comes from one embed token that resolves to exactly one user. ### 2. Read it in four passes **Spend.** Which model was actually reached, how many input and output tokens, what it cost. Note that cost is recorded per request against an identity, not estimated afterwards from a log — that is what makes it chargeable. **Latency.** How long calls took. Say where the time went: provider latency is not the same as governance latency, and the policy chain is synchronous and fast. If the numbers show that, say so; if they don't, say that instead. **Verdicts.** Every allow and every deny, with the deciding policy and reason. Group them: which polici