Agent posture

packs:agents

An agent posture row is one agent enrolled in this estate with everything the platform knows about its standing: tier, owner, scope, release, instances, last action, caps and spend, transcripts, data-handling marker, disclosures, health and staleness. One row per agent, never per instance. Served by the v2 AgentPostureService.ListAgentPosture; reads need read on the project path (404-on-deny) and an estate-wide read needs read on the root (opschain docs agent-posture).

AgentPostureRow

One enrolled agent's posture.

Field	Type	Meaning
agent_id	string	The agent's id: the member id of its most recently registered instance — the id opschain admin agent trust-review and the activity timeline address.
agent_type	string	The declared agent type: service, ai or "" when untyped.
assigned_scopes	string[]	The assets-under-management scopes assigned to the agent's principal, scope ascending.
code	string	The agent's code, unique in its project.
data_handling_approved	boolean	True when the model-provider catalog record the agent brokers through carries the data-handling-approved marker.
disclosures	AgentPostureDisclosures	The manifest's disclosures, summarised.
group	string	The fleet group the agent's current instance belongs to.
health	string	The agent's liveness: connected, offline, stopped, halted or draining.
instances	integer	Instances enrolled (not retired).
instances_connected	integer	Of those, instances holding a live channel right now.
last_action	AgentPostureLastAction	The last governed action the ledger holds; unset when the agent never acted.
last_seen_at	timestamp	When any instance was last seen; unset when never.
model_provider	string	The model-provider catalog record the agent brokers through; "" when none.
name	string	The manifest's display name.
over_cap	boolean	True when any token-cap window's spend has reached its cap in the current period.
owner_departed	boolean	True when an owner is set but the identity catalog no longer holds an enabled user or a known group by that name (the owning human left — an identity sweep leaver — or the team was removed). A posture flag only: tier and caps are unchanged; it clears when the access-review work item reassigns ownership.
owner_principal_id	string	The owning user's principal id when the reference resolves to an enabled user; "" for a group owner, an unset owner or a departed one.
owner_ref	string	The owner reference as the manifest or the admin override recorded it — a username or a group name from the identity catalog; "" when no owner is set. Never exported to CSV.
principal_id	string	The agent's governed principal id: its service-account principal on an identity-backed composition, the synthetic agent:<project>/<code> id otherwise. Grants and assignments are addressed to it.
project_path	string	The project the agent is enrolled in (/code).
release	AgentPostureRelease	The release the agent's current instance enrolled from; unset when it enrolled without one.
stale	boolean	True when no governed action landed within stale_after (never acted included).
tier	string	The effective tier from the compiled trust snapshot: the most permissive tier any active grant confers on the agent's principal, group or attributes, clamped by its type's ceiling; advise when only an assignment and the advise floor reach it; none when nothing does.
token_windows	AgentPostureTokenWindow[]	The token-cap windows active grants set for the agent and the spend against each in its current period, window order.
transcripts_captured	boolean	True when the platform holds captured transcripts (chat or adapter transit) for any of the agent's instances.

AgentPostureRelease

The signed release an instance enrolled from.

Field	Type	Meaning
digest	string	The release artifact's digest — the identity an external inventory keys on.
id	string	The release's id.
name	string	The release's name.
stage	string	The release's current promotion stage.
version	string	The release's version.

AgentPostureLastAction

The last governed action in the audit ledger: a request through the governed lane or a brokered model call.

Field	Type	Meaning
at	timestamp	When the action landed.
correlation_id	string	The correlation chain the action belongs to; "" when the ledger row carries none.
event_type	string	The audit event type the action was read from (info:agent:request, warn:agent:request_refused, info:agent:model_call, warn:agent:model_call_refused).
kind	string	request or model_call.
outcome	string	allowed, refused or served.

AgentPostureTokenWindow

One cap window's standing.

Field	Type	Meaning
cap	integer/string	The cap the grant sets for the window.
grant_id	string	The grant setting the cap.
used	integer/string	Tokens spent in the window's current period.
window	string	minute, day, week, month, quarter or year.

AgentPostureDisclosures

The manifest disclosure summary.

Field	Type	Meaning
adapter	string	The disclosed adapter.
custody_mode	string	The disclosed model-credential custody mode.
model_name	string	The disclosed model name.
model_provider	string	The disclosed model vendor.
model_version	string	The disclosed model version.
purpose	string	The disclosed purpose.
tool_surface_count	integer	How many estate tools the manifest declares.

AgentPostureSummary

The counts over the rows returned (filters applied) — the dashboard panel's figures.

Field	Type	Meaning
by_tier	map	Rows per effective tier (advise, supervised, autonomous, none).
over_cap	integer	Rows with a token-cap window at or over its cap.
owner_departed	integer	Rows whose owner has departed.
stale	integer	Stale rows.
total	integer	Rows returned.
without_owner	integer	Rows with no owner set.

AgentPostureFilter

Narrows the list; every unset field keeps all.

Field	Type	Meaning
agent_type	string	Keep only agents declaring this type (service, ai).
departed_owner	boolean	Keep only agents whose owning principal has departed.
project_code	string	Keep only agents enrolled in this project; "" keeps every project the caller may read (needs read on the root).
stale	boolean	Keep only stale agents.
tier	string	Keep only agents at this effective tier (none, advise, supervised, autonomous).

The response also carries stale_after (the threshold the rows were judged against — agents.stale_after, 168h when absent) and computed_at (the server's clock when the rows were composed).