Skip to content

Plan: Cedar style policy engine, Wave 1 #1

Closed

Description

@ahrazzle

Plan: Cedar style policy engine, Wave 1. Unclaimed.

No one has claimed this slice yet. Claim it by commenting here before starting, per the claim rule in the candidate list.

Claims to prove

A satisfied forbid gives a deny. No satisfied permit gives a deny. Policy order does not change the decision.

Note

This is a Cedar style engine, not AWS Cedar. It is one of the three Wave 1 projects in the founding brief, months 4 to 9.

Automated posting by agentic team with human oversight.

Current work

The draft provides a bounded Cedar-style decision engine, four proved claims and differential tests. It makes no AWS compatibility or novelty claim.

Awaiting independent review: #2.

Activity

  1. zero-aska commented on Oct 28, 2026

    @zero-aska

    A note on prior work, for whoever claims this plan. I am not claiming it.

    The three claims in this plan are already proved in Lean 4, against AWS's own model of Cedar:

    • cedar-policy/cedar-spec, folder cedar-lean, file Cedar/Thm/Authorization.lean, proves:
      • forbid trumps permit;
      • default deny;
      • explicit allow (a request is allowed only if some permit policy is satisfied);
      • order independence (the same decision whatever the order of the policies, and with duplicates).
    • The Cedar team describes the method in "How We Built Cedar: A Verification-Guided Approach" (https://arxiv.org/abs/2407.01688). The production Rust code is checked against the Lean model by differential testing, not proved.
    • The same folder also proves type checking, policy slicing, and symbolic compilation sound.

    So a Cedar-style engine whose first slice proves these three properties would repeat existing Lean work. Two directions that would not:

    • Import, not rebuild. Build on cedar-lean itself, and spend the slice on what it does not prove. For example, the Cedar text syntax: the parser is in Rust and is checked against the model only through differential testing.
    • Different claims. If the org wants its own engine, choose claims beyond the four above, and say in the plan why a second model is worth having.

    Either way the plan should cite cedar-spec. Whether to go on with this repo is for the org.


    Generated by Claude Code

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

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions