Skip to content

P2: Add Own.NET protocol-check gate and task contract example #45

Open

Description

@PhysShell

Type

Enhancement / integration gate.

Priority

P2 / Medium.

Tags

gate, ownnet, protocols, obligations, task-contract, agentic-coding, ci-evidence

Context

007 should not understand Own.NET semantics. Its job is to isolate, run, gate, and harvest. Once Own.NET produces obligation/protocol findings and OwnAudit can render review verdicts, 007 should be able to gate agent runs on those artifacts.

Target story:

Agent refactors a legacy ViewModel.
Agent accidentally moves OnPropertyChanged(Document) before IsLoaded=true.
Own.NET reports OWN201.
OwnAudit renders protocol-review.md + protocol-verdict.json.
007 gate fails and harvests diff + evidence.

That is the useful loop. Not vibes. Not trust. A tiny cage with a verdict, because apparently that is what it takes to keep robots from rearranging the furniture with a flamethrower.

Proposed scope

Add an example gate and task contract shape for protocol checks.

Possible gate step:

[[steps]]
name = "ownnet-protocols"
cmd = "python -m ownlang ownir protocol-facts.json --sarif protocol-findings.sarif"

[[steps]]
name = "ownaudit-protocol-review"
cmd = "python -m protocol.review_cli --findings protocol-findings.json --baseline protocol-baseline.json --out-dir protocol/out --gate-level error"

Exact commands can change once Own.NET/OwnAudit land their slices. This issue is about reserving the integration seam and example workflow.

Acceptance criteria

  • Add an example .007/gate.protocols.toml or extend the existing Own.NET gate example.
  • Add a sample task contract that says:
    • behavior-changing refactors are forbidden unless explicitly requested;
    • new Own.NET protocol violations fail the run;
    • protocol-review artifacts are harvested.
  • Ensure run record captures:
    • protocol findings;
    • protocol review markdown;
    • protocol verdict JSON;
    • diff.patch.
  • Keep 007 as executor only: no semantic interpretation inside 007.

Non-goals

  • Do not implement Own.NET protocol analysis here.
  • Do not implement OwnAudit reporting here.
  • Do not make LLM judgment decide pass/fail for deterministic protocol violations.

Why this matters

This turns the obligation-protocol idea into an agent safety loop: agents can still write code, but they cannot silently introduce project-specific invariant breaks without a deterministic gate catching them. Small cage, fewer escaped raccoons.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions