Agent enablement

Permanent link /agent-toolkit/agent-contract

Using Ethotechnics as an AI Agent

This contract defines the inputs, outputs, and evidence required when an agent applies Ethotechnics. Use it to keep audits, red-teams, and design reviews consistent.

Agent contract

Supported tasks

Use Ethotechnics for structured audits and mechanism-first remediation.

  • Audit a system against STD-01/STD-02 clauses
  • Red-team for contestability, stoppability, and irreversible harm
  • Design assist: map risks to mechanisms and integration steps
  • Policy drafting: translate standards into enforceable requirements

Inputs

Expected inputs

The minimum information an agent needs to form evidence-backed findings.

  • System description with actors, authority boundaries, and action surface
  • Policy or specification text (draft or live)
  • UI flows or interaction maps for affected users
  • Observed evidence: logs, decision records, or appeal metrics

Outputs

Canonical outputs

Define the artifact set so results stay portable across tools.

  • Diagnostic report with violated clauses and evidence notes
  • Finding objects that chain risks to mechanisms
  • Mechanism set with implementation cues and coverage gaps
  • Remediation checklist with owner + timeline

Citations

How to cite standards, mechanisms, and validators

Citations should make evidence traceable and enforceable.

  • Cite standards as STD-01.C# or STD-02.C# with clause IDs.
  • Cite mechanisms as MEC-## with the canonical permalink.
  • Cite validators as VAL-## with output references.
  • Pin versions in reports (e.g., STD-02@0.9, MEC@v1.1.0).

Executable standards

Clause schema for agent-readable obligations

Every clause should be represented as a structured object with evidence requirements.

{
  "clause_id": "STD-01.3.2",
  "standard_id": "STD-01",
  "level": "MUST",
  "condition": "timeout threshold is reached",
  "obligation": "provide a failure response or human escalation",
  "time_bound": "at timeout",
  "evidence_required": ["escalation_policy", "event_log"],
  "failure_modes": ["timeout without escalation"],
  "related_mechanisms": ["MEC-04", "MEC-06"],
  "related_validators": ["VAL-03"]
}

Finding object

Portable diagnostic artifacts

Findings let agents chain risks to mechanisms and remediation.

{
  "finding_id": "FND-2026-041",
  "severity": "high",
  "summary": "Exit flow introduces retention scripts before cancellation.",
  "violated_clauses": ["STD-01.2.3"],
  "impact": "Users cannot exit without added friction.",
  "closure": "Remove retention gate and keep one-step cancellation.",
  "recommended_mechanisms": ["MEC-02"],
  "confidence": 0.83
}

Negative examples

Known bad patterns and false-positive warnings

Use these to avoid superficial compliance and improve detection precision.

  • Non-binding review promises: Says “human review available” without any timeline or authority; violates STD-02.C1/C3.
    False-positive warning: False positives occur when timelines exist but are buried in legalese.
  • Appeals that cannot touch the outcome: Provides an intake form but never connects to a decision reversal or remedy pathway.
    False-positive warning: Legitimate appeals show reversible state changes and audit receipts.
  • Consent without exit ramps: Uses soft opt-outs that keep automation running or ignore revocation.
    False-positive warning: If the exit pauses automation and records the event, it is compliant.

Prompt library

Agent-ready test prompts

Pair these prompts with the prompt packs for consistent diagnostics.

  • Given a system description, enumerate violated clauses with clause IDs.
  • Map each risk to at least one MEC with justification and required artifacts.
  • Rewrite this policy snippet to satisfy STD-02.C3 with time-bound remedies.
Go to prompt packs

Versioning

Diffable, pinned releases

Use versioned releases so agents can compare results over time.

  • Pin analysis to an explicit release: ethotechnics@2026.1 or STD-02@0.9.
  • Surface diffs when clauses change, with deprecated-to-replacement pointers.
  • Include change log references in every report.

Boundaries

What Ethotechnics does not decide

Ethotechnics is procedural, not moral authority.

  • Ethotechnics does not rank social values or resolve political tradeoffs.
  • Ethotechnics does not certify moral intent; it audits controls and evidence.
  • Ethotechnics does not replace governance; it supplies enforceable primitives.

Optional

Governance ontology primitives

These primitives keep symbolic reasoning consistent across systems.

  • Actor
  • Authority
  • Decision
  • State change
  • Reversal
  • Standing
  • Contestability