Skip to content

SKILL.md

Latest commit

 

History

History
179 lines (124 loc) · 3.94 KB

SKILL.md

File metadata and controls

179 lines (124 loc) · 3.94 KB
name verified-task
description Verify any agent output before it is trusted, paid for, or acted on. A universal SAR verification pattern for AI agent workflows.

Verified Task

A universal verification pattern for AI agents.

Verified Task allows any agent to validate its output before it is used, trusted, or paid for — using deterministic verification via SettlementWitness (SAR).

Core Principle

Do not trust outputs by default.

Verify them before:

  • payment
  • execution
  • downstream use
  • decision-making

What This Does

Verified Task provides a simple, standardized way to apply SAR verification to any task.

It:

  • evaluates output against a defined specification
  • returns a deterministic verdict (PASS / FAIL / INDETERMINATE)
  • produces a verifiable receipt (SAR)
  • signals whether it is safe to proceed

When to Use

Use this skill whenever:

  • an agent produces output that must be correct
  • payment depends on task completion
  • results affect decisions or actions
  • trust needs to be explicit and verifiable

Core Execution Flow

  1. Define task specification:

    • what the agent must do
    • what correct output looks like

  2. Generate output:

    • agent performs the task

  3. Verify output:

    • compare output against specification

  4. Receive result:

    • verdict (PASS / FAIL / INDETERMINATE)
    • SAR receipt (verifiable proof)

  5. Decision:

    • PASS → proceed
    • FAIL → reject or retry
    • INDETERMINATE → review or refine

Verification Call

settlement_witness({ task_id: "task-001", agent_id: "0xYourWallet:your-agent", spec: { expected: task_spec }, output: { result: agent_output }, counterparty: counterparty_id })

Endpoint: POST https://defaultverifier.com/settlement-witness

Input Structure

{ "task_id": "unique-task-identifier", "agent_id": "0xYourWallet:your-agent-name", "task_spec": "clear description of expected outcome", "expected_output_format": "format or constraints (optional)", "output": "actual result produced by the agent", "counterparty": "optional wallet or agent identifier" }

Output Structure

{ "verdict": "PASS | FAIL | INDETERMINATE", "safe_to_proceed": true, "sar_receipt": { "receipt_v0_1": { "verdict": "PASS", "confidence": 1.0, "reason_code": "SPEC_MATCH", "receipt_id": "sha256:...", "verifier_kid": "sar-prod-ed25519-02", "sig": "base64url:..." }, "trustscore_update": { "score": 54, "tier": "bronze", "receipts_to_next": 10 } }, "notes": "Output matched specification. Safe to proceed." }

Decision Rules

  • PASS → output is verified and safe to use
  • FAIL → output does not meet specification
  • INDETERMINATE → cannot be confidently verified

Never proceed with payment or critical actions unless verdict = PASS.

Relationship to SettlementWitness

This skill is a usage pattern for SettlementWitness (SAR).

  • SettlementWitness performs deterministic verification
  • Verified Task provides a simple interface for applying it

This skill does not replace SettlementWitness.
It standardizes how agents use it.

Privacy and Safety

  • No private keys, credentials, or sensitive data should be included in task specifications or outputs
  • Verification is stateless — no data is retained after the receipt is issued
  • All verification is deterministic and reproducible

Integration

Can be used with:

  • payment systems (only pay on PASS)
  • agent workflows (gate execution)
  • trust systems (use receipts as evidence)
  • reputation systems (aggregate verified outcomes)
  • skill marketplaces (wrap any existing skill with a verification step)

Outcome

Agents become:

  • safer
  • verifiable
  • auditable
  • trustworthy by default

Keywords

ai-agents, verification, agent-safety, settlement, trust, automation, workflow-validation, workflow-safety, output-validation, deterministic-verification, agent-commerce, trustworthy-agents, sar-protocol