generated: '2026-07-21'
method: searched
source: https://silmaril.dev/docs
description: >-
  Silmaril's classify API does not return RFC 9457 problem+json errors; its
  "error" surface is (1) the firewall verdict/outcome taxonomy returned on every
  classification and (2) typed block exceptions raised by the SDKs in
  enforcement mode. Both are captured here.
verdict:
  prediction:
    field: prediction
    values: [BENIGN, MALICIOUS]
    note: Required backend field used for enforcement.
outcome_field: primary_outcome
outcomes:
- code: benign
  meaning: No harmful firewall outcome detected.
  action: Continue normally.
- code: information_disclosure
  meaning: >-
    Private data, documents, internal context, logs, traces, customer data, SQL
    rows, topology, or similar non-secret sensitive information.
  action: Block or require review before returning private data.
- code: secret_exposure
  meaning: >-
    Credentials, tokens, API keys, cookies, passwords, signing keys, OAuth
    secrets, session material, or webhook secrets.
  action: Redact or suppress secret-bearing content.
- code: control_abuse
  meaning: >-
    Misuse of authorized tools or user privileges to send, change, approve,
    delete, operate, or bypass policy/RBAC without a stronger outcome.
  action: Deny the action and request explicit confirmation.
- code: system_compromise
  meaning: >-
    Privilege escalation, account takeover, hostile integration/plugin takeover,
    persistence, lateral movement, attacker webhook registration, or code/plugin
    execution.
  action: Block, security-log, and escalate.
- code: service_disruption
  meaning: >-
    Downtime, lockout, degradation, alert suppression, destructive loops,
    resource exhaustion, cost spikes, or hidden outage evidence.
  action: Block destructive or disruptive actions.
exceptions:
- name: FirewallBlockedException
  raised: SDK enforcement mode when a call is blocked.
  fields: [score, threshold]
- name: BatchFirewallBlockedException
  raised: Enforcement mode for batch/long-input fanout classification.
  fields: [score, threshold]
notes: >-
  Score threshold decisions are tenant-owned Firewall backend configuration
  (since SDK 0.4.0). Additional recorded detection categories include model
  distillation and NSFW content abuse (per the MCP evidence surface).