Pretorin Compliance

[Recipe scope-artifact-compose (official)] Render a brand-styled, fully-escaped HTML table for a tabular artifact (PPSM, interconnection). Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe scope-q-answer (official)] Run the candidate answer through pretorin.evidence.redact and return cleaned text + counts. Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe ssp-gap-analysis (official)] Compose auditor-facing Markdown evidence from one or more cited document snippets. Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe ssp-gap-analysis (official)] Detect NIST-style control IDs and CCI IDs in the corpus and return cited occurrences grouped by identifier. Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe ssp-gap-analysis (official)] Extract supported document text into a local page/chunk corpus JSON file with file hashes and warnings. Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe ssp-gap-analysis (official)] Search the ingested document corpus and return top cited snippets without loading the full corpus into context. Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe workspace-capture (official)] Compose an auditor-facing Markdown evidence body with provenance for any workspace artifact. Requires an active recipe execution context for this recipe (call start_recipe first).

[Recipe workspace-capture (official)] Run pretorin's secret redactor over the supplied workspace artifact text. Requires an active recipe execution context for this recipe (call start_recipe first).

Record the independent assessment axis only for a Campaign item that has no frozen objective roster. Objective-backed items must instead use record_campaign_objective_disposition exactly once per UUID; their aggregate assessment is derived. grounded requires scoped evidence, gap requires an authoritative control-gap Issue, and needs_input requires a limitation. This does not alter execution or review.

Record retained, remapped, superseded, or needs_input for the frozen evidence artifact/set. The server verifies every evidence ID remains inside the assigned control. Remapping requires a matching typed evidence_link artifact; superseding records replacements without deleting data.

Record one bounded outcome for the exact frozen Issue. The server re-reads the assigned canonical Issue and verifies lifecycle state and evidence before persisting it. This does not approve treatment, accept risk, or void an Issue.

Record one bounded process disposition for one frozen assessment-objective UUID. grounded requires persisted evidence IDs; needs_input requires a concrete limitation. This does not approve the parent control and stores no narrative body.

Deprecated compatibility tool for Campaign records created before coordinator-owned preflight snapshots. Current Campaign worker profiles do not expose this tool: preflight runs once before dispatch and children read the frozen snapshot. Connection, source, status, and search facts are normalized from the server's current local preflight authority and this attempt's recorded reads; caller values cannot manufacture them. Recipe IDs must be trusted built-ins and selected recipes must match an active content pin. This stores bounded decisions and limitations, not evidence bodies. An unavailable source is a limitation and never prevents best-effort assessment.

Record a content-free skipped or terminal-error outcome that no typed writer can observe. The server derives the logical idempotency key from the parent Plan, child Plan, structural version, step ref, and stable artifact identity. Reusing that identity with different content is a hard conflict.

Record the existing platform review job/generation for the assigned child Plan. This is an observer correlation, not a local approval or review decision.

Re-score the risk and trigger a best-effort AI summary refresh. The endpoint always returns 200 with the latest entry — the re-score commits regardless of AI availability. To confirm AI ran, check whether ai_summary_generated_at advanced after the call.

Reject a submitted canonical Issue treatment plan. Runs no readiness checks — an incomplete plan is still rejectable — and leaves gate_status and the Issue's lifecycle_status unchanged.

Remove an inheritance edge, making the control system-specific.

Reopen an approved policy for editing (regress it to draft, clearing approval and bumping the version). Records a monitoring regression event + audit. Use before update_policy_narrative, answering questions, or regenerating when the policy is approved.

Reopen a completed scope for editing (regress it to in_progress). Records a monitoring regression event + audit. Use before update_scope_narrative, answering questions, or regenerating when the scope is completed.

Reopen a reviewed vendor assessment for correction, moving it to in_progress. This revokes any live vendor portal link, resets the AI advisory to not_run, and retains the prior review as a superseded review_revision. AI scoring must be re-run after resubmission, or the next review must set acknowledge_no_ai_review=true. Only reviewed assessments can be reopened. Pass exactly one of vendor_id or vendor_name.

Atomically replace only one framework's policy mappings; unrelated frameworks remain unchanged.

Reopen or update one existing control Issue while evaluating it through the bounded issue-evaluate recipe. This cannot create child Issues or close through generic metadata update. Closure is governed: use verify_issue after treatment or void_issue for a finding that was never valid.

Deprecated alias for resolve_control_issue.

Nonblockingly adopt or resume an exact confirmed Plan-backed Campaign through the shared coordinator. Unconfirmed drafts require explicit start_campaign confirmation; eligible failed items are reopened without rebuilding the frozen hierarchy.

Record a periodic authorized review of an APPROVED CMMC Operational Plan of Action. Only valid on an approved plan whose kind is cmmc_opa.

Record an explicit completed review of one risk and reset its review cadence. Use only after inspecting the risk, its linked controls/evidence, scoring, and treatment.

Finalize a vendor assessment with reviewer residual likelihood and impact. Set acknowledge_no_ai_review when finalizing without a successful AI advisory summary.

Withdraw an active Issue risk acceptance. Unlike the other governed writes this has no Issue-lifecycle restriction — it works on a verification_pending, closed, or voided Issue — and it can demote the control's approval gate. Only an active acceptance is revocable.

Upsert answer responses for a vendor assessment.

Schedule a formal assessment against the active system/framework so it can be started and displayed in the Auditor Portal. The scheduled end date must be on or after the scheduled start date.

Run advisory AI scoring over a submitted vendor assessment.

Search evidence within exactly one active system/framework scope. With query, performs RAG semantic search over attached evidence and scoped unattached evidence, including org policy documents, so agents can reuse existing evidence before creating new artifacts. Without query, lists evidence linked to control_id or the current framework scope.

Search the product-facing Pretorin Platform Capability Index for features that can satisfy a compliance requirement. Call this before creating local tracking documents, spreadsheets, registers, or placeholder artifacts; if Pretorin has a matching platform surface, use that surface as the system of record. Unauthenticated.

Search one org policy's generated narrative for a natural-language concept. Use this as the supported first step for 'does this policy cover X?' questions before reading every section or editing the policy. Returns ranked section matches with stable section IDs, titles, similarity scores, and short excerpts; an empty matches list means no section met the requested similarity threshold.

Bulk-seed risks from org-level library templates. Auto-links controls per template's suggested_control_families when the system has ControlImplementation rows for the given framework. Returns the seeded entries; raises 400 if no template ids resolve.

Edit the active recipe set for the active (or given) scope and persist it on the preflight artifact. mode='replace' (default) sets the active set to exactly recipe_ids; 'add' activates them and re-pins reviewed same-version, same-source content drift; 'remove' deactivates them. Seed the set from get_active_recipes' candidates during preflight, then adjust as the effort evolves. Ids not in the cookbook are skipped and reported. Establishing an active set makes it the confirmed menu: start_recipe then refuses recipes outside it (unless force=true).

Create or update an inheritance edge for a control. Set responsibility_mode to 'inherited' or 'shared', with source_type and optional vendor_id.

Replace the Issue-owned facts consumed by formal POA&M projections.

Declare or clear the system's target scale tier for a framework scope. The response includes the declared, derived, and effective tiers; pass null to clear the override and fall back to the derived tier.

Transition a vendor to a new lifecycle status (onboarding, active, inactive). Requires the server-side `vendor.pii` scope (or an admin token); the platform returns 403 if the token is not authorized. A non-empty reason (<=500 chars) is required.

Sign authorizing-official acceptance of a vendor's residual risk for one attached system. Idempotent per (vendor, system). Requires the org's evidence attestation-envelope capability and attestation process mode; when signing is disabled the tool returns a clear error.

Start a scheduled formal assessment and queue its immutable auditor snapshot. Starting before its scheduled start date is rejected. Replay is a no-op once the snapshot is freezing or frozen; use get_assessment to poll.

Confirm an exact frozen Campaign preview by digest and start its detached shared coordinator. Returns immediately; use get_campaign or the informational same-host CLI watch command to observe it.

Open a recipe execution context. Returns a context_id the caller passes on subsequent platform-API write tool calls so audit metadata is stamped with producer_kind='recipe' automatically. Pass control_id to pin the context to a single control; OMIT control_id to open a FAMILY (multi-control) context scoped to system_id + framework_id — one family context stamps writes for every control in that scope, so a family/campaign run can author N controls without reopening a context per control (each write still names its own control_id). The response 'scope' field reports 'control' or 'family'. One recipe per session at a time (nesting forbidden in v1). Contexts auto-expire after 1 hour of inactivity. When executing a declared Plan recipe step, pass plan_id and step_index; the step must be in_progress and its recipe id/version must match. Only the resulting start/end receipt permits that step to complete. Opening a context is a platform write, so it is refused when the resolved system/framework doesn't match the active context; switch the active context with `pretorin context set` to target another scope. It is also refused when a source kind the recipe requires is verified missing in the local preflight verdict, or — on a provisioned scope — when the recipe is not in the active set or its pinned version/content/source has drifted (re-activate same-version/same-source content; replace a changed version/source pin; or pass force=true to run it ad hoc).

Route a user prompt to the right workflow. Call this after check_context whenever the user references compliance work (a control, system, framework, questionnaire, source preflight, or campaign). The calling agent extracts entities from the user prompt and supplies them as structured args; pretorin applies deterministic rules to pick a workflow and bundles the platform read-state (workflow_state, compliance_status, pending items) into the response. The agent then reads the selected workflow's body via get_workflow and follows it. ARGUMENT SHAPE: all prompt-derived fields (intent_verb, raw_prompt, system_id, framework_id, control_ids, scope_question_ids, policy_id, policy_question_ids) MUST be nested inside the `entities` object. Only `active_system_id`, `active_framework_id`, `skip_inspect`, `create_new_plan`, `parent_plan_id`, and `campaign_context` live at the top level — those come from the CLI runtime, not the user prompt. Flattening prompt entities to the top level is a common caller bug: such top-level copies are ignored by the handler (not read, not rejected), so the route is decided from `entities` alone and the misplaced fields silently have no effect.

Submit an asset-inventory diff produced by a recipe-driven scan. At least one of `added` / `modified` / `decommissioned` must be non-empty. `recipe_id` is a free-form CLI string the platform records as `cli:<recipe_id>` provenance per row. `idempotency_key` defaults to sha256(system_id, recipe_id, scan_timestamp); pass your own when replaying a scan. `recipe_context_id` (from start_recipe) is optional — the diff endpoint accepts it cleanly but does not require it.

Deprecated checkpoint compatibility only. Persist one external-agent proposal on a legacy prepared item without applying it. Plan-backed workers publish through typed MCP writers.

Submit a complete canonical Issue treatment plan for approval review. The reviewer then calls approve_issue_plan or reject_issue_plan.

Submit a ready policy for first-party human review. This cannot approve a policy.