{{ crumb.title }}
Safety alerts
Risks that need attentionTrust score
Success Rate
Activity
Success Rate
Checkpoints
Latest checkpoint resultsQuality checks
Test results + evidenceRecent tasks
Recent activity
Data-leak risks
{{ dash.trifectaNote }}
Recent activity
Recent activity


{{ iss.title }}
{{ b.h }}
{{ b.p }}
{{ b.code }}{{ b.h2 }}
{{ b.h }}
{{ b.p }}
{{ b.code }}
{{ ag.name }}
{{ b.h2 }}
{{ b.h }}
{{ b.p }}
Autonomy envelope
{{ run.env.lead }}
{{ run.env.cause }}
Per call, the gate also checks what a tier cannot show — e.g. money above a threshold or a tool outside the agent's declared purpose.
Tiers are the tool's own declaration: read-only, internal write, external send. No scores are computed.
Governance timeline
Standing rules for which calls wait for a person. The gate reads them on every call; a rule never widens an agent's tools.
Work an agent marked done that a person or a verifier agent checks before it counts.
Routines
Recurring work that runs on its own and creates an Issue you can track each time.
{{ prj.name }}
{{ prj.desc }}
Work queues
Concurrent lanes for parallel agent work. Atomic claim, retry, dead-letter.
{{ b.h2 }}
{{ b.h }}
{{ b.p }}
{{ b.code }}Learned playbooks
Patterns distilled from successful runs. Approve to keep one, then create a skill from it; reject to dismiss.
Trigger synthesis from a successful run to see candidates.
{{ b.h2 }}
{{ b.h }}
{{ b.p }}
Costs
Inference spend, platform fees, credits, and live quota windows.
Inference ledger
By agent
By model
Finance ledger
Soft alert at 80%. At 100% new runs wait for a person; nothing already running is cut off mid-task.
{{ set.sel.label }}
{{ set.sel.desc }}
Reuse a governance posture — autonomy level, allowed tools and deny floor — under a name and attach it to a scope.

{{ set.sel.note }}