auditlisted

You are the audit query skill. When invoked, you produce a concise summary of every tool call recorded in the current session's audit ledger. The ledger is written by the audit-toolkit hook (`audit-tool-calls.sh`) on `PostToolUse` / `PostToolUseFailure` / `SubagentStop` / `UserPromptExpansion`. The hook fires automatically once the `audit-toolkit` plugin is enabled (plugin `hooks/hooks.json`); a legacy manual mode wires it via `.claude/settings.local.json`. It is the harness's record of what tools fired AND which slash commands the user invoked, independent of Claude's chat output. **When the two streams disagree, the harness wins.** This skill is **observability only**. It surfaces signals for manual review and team-convention checks; nothing here blocks pushes, commits, or any other action. ## Capabilities - ✅ Surfaces every tool call recorded since session start. - ✅ Distinguishes successful (`outcome=ok`) from failed (`outcome=fail`) calls. - ✅ Surfaces user-typed slash-command loads (`UserPromptExpansion` events). - ❌ No hash chain or tamper detection (lean version has none). - ❌ Does NOT block pushes / PR creation. This is a visibility signal, not a gate. - ❌ Does NOT see `Skill()` tool invocations (Claude Code does not fire `PostToolUse` for them — `/dev-pipeline` nested loads are invisible to the audit by design). ## Inputs - **Optional**: `--session-id <ID>` — query a specific session's ledger instead of the current one. - **Default**: query the most-recently-m