Module trust_report 

Opt-in periodic trust report.

A digest of what the agent actually did over the reporting window — task runs, completion-contract verdicts, token spend, guardrail interventions, delivery/recovery health — delivered to every control chat. Rendering is a pure function over a snapshot struct; the loop only gathers, formats, delivers, and records the last-sent marker in db_meta so cadence survives restarts. No LLM calls are involved.

Structs

TrustReportData

Everything one report renders, gathered up front so formatting is pure.

Functions

format_trust_report

report_due

True when the persisted last-sent marker is at least interval_days old (or absent/unparseable — a fresh install reports on the first tick).

run_trust_report_once

One cadence check: send the report to every control chat when due. Returns true when a report was delivered to at least one chat.

spawn_trust_report