Skip to content

Provenance-aware authority flow protection - #53

Open
markndg wants to merge 1 commit into
mainfrom
feature/provenance-authority-flow
Open

Provenance-aware authority flow protection#53
markndg wants to merge 1 commit into
mainfrom
feature/provenance-authority-flow

Conversation

@markndg

@markndg markndg commented Aug 24, 2026

Copy link
Copy Markdown
Owner

Summary

  • Adds Provenance-Aware Authority Flow: explicit provenance sources, taint lattice, and the capability-vs-delegation distinction (§8–9) so untrusted causal chains cannot exercise privileged authorities the agent merely possesses.
  • Hooks enrichment into enrich_action before PolicyEngine.evaluate; persists sources/edges/delegations/findings/tool fingerprints (DB migration v8).
  • Ships provenance-authority-defense policy pack, /ui/authority dashboard, CLI (provenance demo|evaluate|explain, authority violations), SDK observe_provenance / provenance_scope, docs, and evaluation corpus.

Test plan

  • Baseline before change: 266 passed
  • Full suite after: 289 passed
  • tests/provenance: 25 passed (core + adversarial + API)
  • varden provenance evaluate: recall 1.0, precision 1.0, FPR 0.0, p95 ~0.17ms
  • Frontend build succeeds with Authority & Provenance page
  • Manual: varden provenance demo and inspect /ui/authority

Made with Cursor

Introduce an explicit provenance + taint + delegation model so untrusted
content cannot acquire authority merely by being read. Guarding now
enriches every action with required vs delegated authority before
PolicyEngine evaluation, with fail-closed pack, APIs, CLI, dashboard,
corpus evaluation, and adversarial/property tests. Client trust claims
remain untrusted unless verified by the control plane.

Co-authored-by: Cursor <cursoragent@cursor.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant