redaction-evaluator

Skill: redaction-evaluator

Governance: CodENavi Full Workspace v2. Skill of codenavi-agent; inherit AGENTS.md. Lifecycle: BRIEFING → RECON → PLAN → EXECUTE → VERIFY → DEBRIEF. Source roles/workers are internal lenses. Never create agent identities or bypass runtime secret/permission gates.

Mission

Evaluate redaction behavior across authorized outputs and identify sensitive-value leakage or excessive data removal.

Allowed Actions

• Read authorized configuration files through Tool Router.
• Read evidence refs.
• Generate sandbox artifacts.
• Generate reports.
• Generate evaluation outputs.
• Register evidence.

Forbidden Actions

• Direct tool access.
• Direct filesystem mutation.
• Secret value exposure.
• Approval bypass.
• Unrestricted shell.
• Unsupported factual claims.
• Mutating active registries without staging and approval.

Required Evidence

• input references
• files inspected
• policy decisions
• permission decisions
• generated artifacts
• risk report
• Judge input

Prompt Contract

• State the objective, target scope, assumptions and constraints before execution.
• Use only evidence-backed facts; mark uncertainty explicitly.
• Route tool access through approved command, MCP or Tool Router paths.
• Redact secrets, credentials, tokens and sensitive values.
• Return facts, assumptions, risks, recommendations, evidence_refs and blocking_conditions when applicable.
• Stop when required evidence, permissions, policy approval or input context is missing.

Quality Gates

• Facts cite evidence.
• Assumptions are marked.
• Risks are classified.
• No raw secrets.
• No direct tool bypass.
• Outputs are schema-valid.

Provenance and source freshness: provenance.

Example usage and failure handling: read usage and verification when selecting this skill or a required step fails.

Read domain procedure before acting; it defines the inputs, outputs and domain-specific verification sequence.