Screenata

Get full policy content including claims and linked test counts. Use the policy slug (e.g., "incident-response"). Set include_content=false to skip the full document.

Enqueue a deterministic draft of the policy from its recorded answers. Blocked (returns the unanswered questions) until every question is answered, unless force=true carries an explicit "draft with what we have". Blocked with reason replace_required on an approved or imported policy, because the draft replaces its current content, unless replace=true carries the user's explicit agreement. Never approves; the draft is reviewed and approved in the portal.

List the questions a policy is built from (the control blueprints citing it): each with its options, the recommended option, the framework requirement it serves, the claim it produces, and whether it is already answered. Use FIRST when drafting a policy: ask only the unanswered ones, at most 4 per turn, quoting controlRef as the reason.

List risk register items. Filter by rating (HIGH, CRITICAL, MEDIUM, LOW) and status (OPEN, CLOSED).

Trigger a cloud or code scan. Provider must be connected in Settings. Supports dry_run for preview.

Use this when the user asks how their compliance program or a framework is doing overall. Returns the workspace readiness score, control coverage, test and policy approval, evidence and open risks, plus one row per enrolled framework: its open audit cycle with the cycle readinessScore (the score shown on the audit cycles page), tests and controls. Pass cycle to pin the header to one audit cycle. Not for automated infrastructure results (screenata_assess) or audit blockers (screenata_audit_check).

Submit text evidence for a test. Provide the test ID (or task code), a title, and the evidence content. Supports dry_run for preview. Binary files (screenshots, PDFs) are CLI-only.

Get one test with why it fails and how to judge it: terminalPath, applyPath (screenata | agent | manual — who applies the fix, decided from the remediation registry and the integration access level; never offer a Screenata apply otherwise) with applyReason and judgmentKit.manualSteps, lastRun (latest check result, errorMessage, failing resources with remediation, remediationSteps), findings (open RemediationTickets with condition / criteria / closureCriteria) and judgmentKit (instructions, requiredElements, auditorAsk, auditorGuidance). Read this before proposing a fix.

Apply a prepared plan after the user explicitly approved it in this conversation. Pass the plan verbatim and the planHash from prepare; a mismatch is rejected. Records a RemediationExecution with before-snapshot and undo, then re-runs the check as its validation run.

Prepare a native remediation plan for a failing check-backed test: steps, current and proposed state, side effects, reversibility and planHash. Show every step to the user; nothing is applied.

Roll back the most recent applied remediation on a test from its before-snapshot.

Re-run the native check bound to a test on the server. The only way a fix becomes a pass: a PASS closes the open finding through post-verification; no status is written directly. Set wait=true to block until the run finishes (up to ~60s) and get the result inline.

Get one run of a test: phase, resultStatus, errorMessage, failing resources and the finding it verified or reopened.

List tests with their status and terminalPath (which fix shape applies from a terminal). Set fixable=true to keep only tests an agent can act on, ranked rerun → export → record → document → integration_blocked.

Who this credential acts as and where: user, workspace (id, slug, title), organization, credential type and scopes. Call it first and name the workspace back to the user before any write.

Use this when answering a question about the connected workspace's compliance policies, procedures or evidence. Search its internal documentation for relevant passages. Not for general framework explanations or unrelated writing. Do not request patient records, payment card data, government identifiers or access credentials.