Both endpoints require Bearer token authentication.
POST /v1/policy/check
Evaluate an agentic action against the workspace policy and AgentGuard rules. Returns a structured decision (allow / deny / confirm / approve) with the factors that drove it and
the residual cost impact.
Use this before any write, deploy, or irreversible action so the agent can self-gate rather than
relying on human review after the fact.
POST
https://api.prysm1.com/v1/policy/check · Auth requiredRequest Body
string
required
The workspace whose policy to evaluate against.
string
required
Natural-language description of the action (e.g.
"deploy to production", "delete user records").string
Structured action type:
edit | refactor | deploy | review | setup | delete.
Influences risk scoring independent of the free-text action field.string[]
Paths that will be read or written. Cross-checked against
protected_paths in the workspace manifest.string[]
Shell commands the agent intends to run. Cross-checked against
forbidden_commands.number
Agent’s estimate of the cost of this step. Used for budget-cap evaluation.
object
Arbitrary JSON context passed through to the decision record (useful for linking to handoff IDs,
session IDs, or external ticket references).
Response
string
allow | deny | confirm | approve. Maps directly to AgentGuard risk tiers:
auto → allow, confirm → confirm, approve → approve, block → deny.integer
0–100 composite risk score.
string
auto | confirm | approve | blockobject[]
Ordered list of contributing factors:
id, label, weight, triggered (bool).string
Human-readable explanation of the decision.
string
The audit event ID for this check (
ae_...).string
ISO 8601 timestamp.
GET /v1/audit/events
Paginated, filterable log of every policy check, handoff, and AgentGuard decision in a workspace. Immutable — entries are append-only and cannot be deleted via the API.GET
https://api.prysm1.com/v1/audit/events · Auth requiredQuery Parameters
string
Filter to a specific workspace. Omit to return events across all your workspaces.
string
Filter by event type:
policy_check | handoff_created | agent_plan | agent_record.string
Filter policy_check events by decision:
allow | deny | confirm | approve.string
ISO 8601 datetime — return events created at or after this time.
string
ISO 8601 datetime — return events created before this time.
integer
Max results per page (1–200; default 50).
string
Pagination cursor from
next_cursor in a prior response.Response
string
audit.eventsinteger
Number of events in this page.
boolean
Whether more pages exist.
string | null
Pass as
cursor in the next request.object[]
Array of event objects. Each has
id, event_type, workspace_id, actor, payload,
created_at. Policy check events also include decision, risk_score, risk_tier.