aiignore
A testable policy proposal for controlling how agentic harnesses access files, environment variables, networks, and sensitive strings.
Experimental draft 0.1. This is not an industry standard, has not been independently audited, and is not a security boundary by itself. Assurance depends on the enforcement paths a harness actually mediates.
Start with one file
Structured policies use .aiignore.yaml. The exact filename .aiignore remains reserved for tools that already interpret it as gitignore-style path syntax.
aiignore: "0.1"
defaults:
network: deny
rules:
files:
- id: secrets
effect: deny
paths: ["**/.env*", "secrets/**"]
Normative artifacts
- Draft specification
- Closed protocol-token registries
- Authoritative errata index
- Policy JSON Schema
- Portable decision schema
- Secret-safe audit-event schema
- Secret-safe readiness-report schema
- Implementation-conformance report schema
- Conformance-report schema
- Detached report-signature schema
- Decision-vector schema
- Parser-vector schema
- Harness-vector schema
- Artifact-manifest schema
- Requirements-traceability schema
- Versioned conformance manifest
- Normative requirements traceability
- Baseline decision vectors
- Security decision vectors
- Option-coverage vectors
- Portable limit vectors
- Parser vectors
- Codex sandbox harness plan
- Published artifact checksums
- Source repository and RFC discussions