agent-session-auditlisted

# Agent Session Audit ## Scope and privacy Use this Skill for an operational read of local Agent sessions: spend, token burn, cache use, tool failures, retry loops, latency, health, anomalies, diffs, and CI gate readiness. Session prompts, code, paths, and outputs may be private; keep analysis local and never upload raw logs to an external service. This audit measures observability signals, not the truth or quality of the Agent's conclusions. ## Discover before interpreting Start with discovery unless the operator supplied a specific export or session directory. Identify the parser/CLI and its version, detected data roots, file format, time range, redaction status, and session capability: - `Detailed`: event/step-level records support failure, retry, and latency analysis; - `Aggregate`: summaries support totals and trends but not event-level causality; - `Limited`: partial metadata supports only the stated fields. If no session data is detected, run the parser's diagnostic command if available and report detected locations plus the next collection step. Do not treat “no data” as zero cost, zero failures, or healthy behavior. ## Audit workflow 1. Preserve the supplied path or discover local exports without changing them. Confirm scope and time window. 2. Run the approved local parser in a human-readable format for investigation and JSON/structured output for automation. Keep output paths user-owned; do not overwrite existing reports unless authorized. 3. Check parser h