Add decision-os-min to Agent Governance & Policy Enforcement

[Aliipou]

Relevance

decision-os-min is a Python execution-governance runtime: a kernel signs an action-bound decision, mints a one-time capability, and a PEP will not run the tool without both. OPA/Cedar stay replaceable grant/deny PDPs. Primary category: Agent Governance & Policy Enforcement.

Repo: https://github.com/Aliipou/decision-os-min

Quality evidence (under 5 stars)

This is an early public reference implementation. Stronger evidence than stars:

• Public, maintained repository (commits within 12 months)
• In-repo destructor / red-team regressions for sealed executor substitution, decide() snapshot, ticket consume-after-verify, dual-PDP mint, audit-before-effect
• Optional official OPA HTTP and Cedar as PDPs only; kernel still signs and spends
• Comparison notes vs OPA/Cedar/MCP are labeled evidence, not proofs
• Explainer: https://ali-decision-os-min.vercel.app (static walkthrough; it does not sign or execute)

Advisory signals (honest)

• GitHub star count is currently below the 5-star advisory threshold
• License is PolyForm Noncommercial 1.0.0 (source-available, not OSI). GitHub reports it as Other. I am not requesting a silent exception file edit; maintainers should reject if the list requires OSI-only software.
• Not a standard, not production-grade, not an AGI-safety proof. TM-A full isolation remains PARTIAL.

Author disclosure

I am the author (Ali Pourrahim / @Aliipou). One project, one category, alphabetical by displayed name (decision-os-min after DashClaw, before defenseclaw).

[CodeSigils]

Thanks for the focused submission. The category, ordering, and relevance look good. We cannot merge this currently because the project is software under PolyForm Noncommercial rather than an OSI-approved open-source license, and no applicable maintainer exception was requested.\n\nSeparately, this PR currently has no completed CI checks reported. The documented awesome-lint and validate-repos checks should be present before any future reconsideration, but missing checks are not the licensing decision. Please revisit if the licensing changes or the project becomes eligible under an explicitly approved policy exception; otherwise we will decline this addition.

[CodeSigils]

This request may be reopened if the license changes or an approved exception is provided

[Aliipou]

@CodeSigils the license call is fair.

decision-os-min is PolyForm Noncommercial 1.0.0 by design. It is not OSI-approved, and I will not rephrase it as if it were. Your category/ordering check is noted. I will not ask for an exception on this PR.

If the license ever changes to something your list accepts, I will reopen with lint/CI green. Until then, decline is the right outcome.

[Aliipou]

@CodeSigils the license call is fair.

decision-os-min is PolyForm Noncommercial 1.0.0 by design. It is not OSI-approved, and I will not rephrase it as if it were. Your category/ordering check is noted. I will not ask for an exception on this PR.

If the license ever changes to something your list accepts, I will reopen with lint/CI green. Until then, decline is the right outcome.

[CodeSigils]

+1 on the decline for licensing. PolyForm Noncommercial is a valid license, but it's not OSI-approved, and this list requires open-source for software entries — no exception should be made here.

Credit to @Aliipou for the honest disclosure in the PR body. That's the right approach.

If decision-os-min ever moves to an OSI-approved license (MIT, Apache 2.0, etc.), a fresh PR with green CI would be welcome. The project itself sounds like a solid fit for the Agent Governance category — the signed-action + capability-spending model is a distinct approach worth tracking.

[ozereray]

This is one of the more interesting execution-governance patterns I've seen recently.

The action-bound decision + one-time capability + PEP requirement creates a useful invariant: the tool cannot execute unless the authorization artifact is actually present at the execution point.

That distinction matters because I think there are really four different states worth keeping separate:

user authorization -> agent proposed action -> policy decision -> actual execution.

At Aegisora we're exploring a similar execution-boundary model with explicit ALLOW / BLOCK / ESCALATE decisions and evidence binding around the agent, tool, decision and execution.

The part I'd be most interested in comparing is the lifecycle of the capability after the decision: how do you prevent replay or substitution when an agent retries, delegates, or changes the target between decision time and execution time?

Nice to see OPA/Cedar kept replaceable as PDPs while the execution kernel remains authoritative.

[CodeSigils]

Closing this PR formally, per the discussion above.

Decision: decline — PolyForm Noncommercial 1.0.0 is not an OSI-approved open-source license, and per CRITERIA.md/contributing.md, software entries require one. This is a policy gate, not a judgment on the work: the signed action-bound decision + one-time capability + replaceable OPA/Cedar PDP model is a distinct and interesting approach in the Agent Governance space, and the disclosure in this PR was exemplary.

Reopen path: if decision-os-min moves to an OSI-approved license, a fresh PR with the repo's CI checks passing would be welcome.

Thanks again for the honest conversation, @Aliipou — and to @ozereray for the technical discussion.