Tool call ledger

What this page covers

The ledger records each managed AI agent action. Task details cover every call in the task, while session details cover only calls in that session. This connects approved scope, actual work, and the agent’s own report.

How to set it up

In task details, review the requester and owner, per-item approvals, sessions, and reports before opening the ledger. Rows show time, tool, decision such as allowed or denied, reason, and result. A blocked command’s input appears in the ledger and can be checked against the recording; because it never reached the target, it does not enter ordinary command records.

Sensitive fragments in readable parameters use a fixed mask, while originals are encrypted. An authorized reviewer enters a reason under Sensitive content access; the access is audited before the original is delivered. Capability handles and credentials are stored only as irreversible fingerprints. Evidence bundles that include operation logs can contain masked agent_tool_calls.json data.

What auditors can see

The ledger preserves readable projections of per-call decisions, reasons, parameters, and results, including denials and blocks. The recording is the source of truth for text sessions; input without echo leaves no command-text record. The query console’s structured statement record is the source of truth for that query. Access to sensitive originals has a separate audit record with the reason and actor.