openvang · matedev01 · Aug 7, 2026 · Aug 7, 2026
@@ -1,2 +1,2 @@
-# Default reviewer for all changes. Routes review requests; see REVIEW.md for the rubric.
+# Default reviewer for all changes. Routes review requests to the owner workflow.
 * @matedev01
@@ -1,10 +1,7 @@
-> ⛔ **Target the `test` branch, not `main`.** PRs into `main` from anywhere but `test` are auto-rejected — see [CONTRIBUTING → Branches](../CONTRIBUTING.md#branches).
+> ⛔ **Target the `test` branch, not `main`.** PRs into `main` from anywhere but `test` are auto-rejected.
 
-> **Check your contribution route before opening this PR.** Contributor PRs may directly change
-> only `agent/**` or `agent.py`, with optional companion `tests/**`. A tests-only PR or any change
-> to another path—including a mixed agent/non-agent PR—requires an open linked issue carrying
-> `benchmark-change-approved` before the PR is opened. See
-> [CONTRIBUTING → Agent submissions and protected project changes](../CONTRIBUTING.md#agent-submissions-and-protected-project-changes).
+> **Protected change route.** Changes outside `agent/**`, `agent.py`, and companion tests require
+> an open linked issue carrying `benchmark-change-approved` before review.
 
 ## Summary
 
@@ -13,7 +10,7 @@
 ## Related issue
 
 <!-- e.g. Fixes #123 -->
-<!-- For any protected or mixed-surface contributor PR, the OPEN issue above must already carry benchmark-change-approved. -->
+<!-- For any protected or mixed-surface change, the open issue above must already carry benchmark-change-approved. -->
 
 ## Type of change
 
@@ -41,5 +38,5 @@
 - [ ] `VANGUARSTEW_OFFLINE=1 python -m pytest -q` passes
 - [ ] Added/updated tests for the change
 - [ ] Updated docs (README / ROADMAP / CHANGELOG) if needed
-- [ ] My changed-file set is an agent submission, or the linked open issue was approved before this PR
+- [ ] My changed-file set is in the component surface, or the linked open issue was approved before this PR
 - [ ] No secrets, tokens, or private data included
@@ -8,7 +8,7 @@ name: Agent benchmark smoke
 #
 # The real score-delta that decides a perf:* band is a maintainer-bot-run LIVE benchmark
 # comparison against BOTH the public curated set and a private, undisclosed repo set
-# (scripts/score_pr_delta.py + combine_dual_target(), see REVIEW.md § Contribution value
+# (scripts/score_pr_delta.py + combine_dual_target(), see benchmark policy
 # labels). That step needs a funded model key and a curated hidden set, so it is not run
 # unattended on every push.
 
@@ -73,7 +73,7 @@ jobs:
               f"- Composite deltas: `{json.dumps(report.get('composite_deltas'))}`\n\n"
               "This does not determine a perf:* label. That requires a maintainer-bot-run "
               "live benchmark comparison against both the public and private repo targets "
-              "— see REVIEW.md."
+              "— see the benchmark policy."
           )
           open("/tmp/comment.md", "w").write(body)
           PY

@@ -1,6 +1,6 @@
 name: PR limit
 
-# Enforce the per-contributor open-PR limit by auto-closing the excess. Runs when a PR is
+# Enforce the per-author open-PR limit by auto-closing the excess. Runs when a PR is
 # opened/reopened (closes it if it puts the author over the limit) and on a periodic sweep
 # (closes any lingering excess). The maintainer is exempt. Uses pull_request_target so it has a
 # write token for fork PRs; it never checks out PR code, so there is no code-execution risk.

@@ -23,5 +23,5 @@ jobs:
           PR_NUMBER: ${{ github.event.number }}
         run: |
           gh pr close "$PR_NUMBER" \
-            --comment "Closing: this PR targets \`main\` directly. Contributors must open PRs against the \`test\` branch per [CONTRIBUTING.md](https://github.com/gittensor-vanguard/vanguarstew/blob/main/CONTRIBUTING.md#branches). Please open a new PR against \`test\`. Thanks!"
-          echo "::notice::Closed PR #$PR_NUMBER — contributors must target test, not main"
+            --comment "Closing: this PR targets \`main\` directly. Please open a new PR against \`test\`. Thanks!"
+          echo "::notice::Closed PR #$PR_NUMBER — changes must target test, not main"
@@ -1,52 +1,44 @@
-# vanguarstew — project constitution
-
-Durable project-wide rules. Every agent, contributor, and CI check operates under these.
-Edit this file when policy changes; code, tests, and CI follow.
-
-## Agent contract (M0)
-
-- The system shall expose `solve(repo_path, request, ...)` as the single entrypoint.
-- The system shall accept `api_base`, `api_key`, and `model` as managed-inference parameters.
-- WHEN `VANGUARSTEW_OFFLINE=1` THE system SHALL use a deterministic offline stub.
-- Agent files are declared in `vanguarstew_agent_files.json` — only those files are scored.
-
-## Benchmark integrity (M1–M3)
-
-- IF the LLM emits a non-string field where a string is expected THEN the scoring pipeline SHALL coerce and log a warning, not crash.
-- IF a repo contributes zero tasks to a multi-repo composite THEN the system SHALL exclude it from aggregation.
-- The system SHALL NOT let a forward-looking signal leak through the freeze boundary.
-- Held-out repos SHALL be scored in a separate generalization pass, not in the tuned pass.
-
-## Code quality
-
-- The system shall reject PRs that lower test coverage below 75%.
-- WHEN code changes under `agent/` or `benchmark/` THEN the PR SHALL include or update tests under `tests/`.
-- `ruff check .` and `VANGUARSTEW_OFFLINE=1 python -m pytest -q` SHALL pass before merge.
-
-## Contributors
-
-- WHILE a contributor has >2 open PRs THEN CI SHALL block new PRs from that author.
-- WHEN a contributor opens a PR against `main` THEN CI SHALL auto-close with a test-branch redirect.
-- A contributor PR confined to `agent/**` or `agent.py` MAY include companion `tests/**` changes
-  and SHALL enter the agent benchmark and Polaris TEE verification route without guardrail
-  preapproval.
-- Every contributor PR with any changed path outside that agent submission surface, including
-  `.github/**`, SHALL require a linked open issue carrying `benchmark-change-approved`; CI SHALL
-  auto-close it otherwise. Adding an agent file SHALL NOT exempt a mixed-surface PR.
-- PRs SHALL reference at least one issue (e.g. `Fixes #N`).
-- Commits SHALL NOT carry AI co-authorship or attribution markers.
-- Contributors SHALL target the `test` branch. The maintainer promotes `test` → `main`.
-- WHEN a closed PR is reopened by an actor other than `matedev01` or `vanguarstew` THEN CI
-  SHALL re-close it; contributors SHALL ask a maintainer to reopen a corrected PR.
-- IF Git metadata claims the contributor PR author's account name for a commit role but GitHub
-  attributes that author or committer role to a different account, THEN CI SHALL close the PR on
-  each PR update and after every CI completion.
-
-## Scoring (gittensor SN74)
-
-- `perf:*` labels, earned only from a measured benchmark delta, SHALL be the sole source of
-  multiplier tiers for `agent/` PRs. Every other surface SHALL carry the flat
-  `mult:contribution`. An unlabeled merged PR earns zero (`default_label_multiplier` is `0.0`).
-- The subnet's `master_repositories.json` entry for this repo SHALL be the authority for every
-  multiplier value; the docs mirror it and lose to it on any disagreement.
-- The 3-axis rubric (repo, maintainer, legibility) SHALL feed into emission weight.
+# OpenVang project constitution
+
+Durable project-wide rules for the OpenVang agent factory and the Vanguarstew
+maintainer-intelligence component. Code, tests, and automation must follow
+these rules.
+
+## Component contract
+
+- The maintainer component shall expose `solve(repo_path, request, ...)` as its
+  stable entrypoint.
+- Managed inference parameters are supplied by the controller; agent code shall
+  not discover or substitute credentials.
+- `VANGUARSTEW_OFFLINE=1` shall select the deterministic offline stub.
+- Benchmark and live persistent memory shall remain time-safe, controller-owned,
+  and read-only from the maintainer component.
+
+## Benchmark and execution integrity
+
+- Forward-looking signal shall not cross a benchmark freeze boundary.
+- Held-out repositories shall be evaluated separately from tuned repositories.
+- Public artifacts and TEE evidence shall contain only receipt-safe commitments,
+  never raw private memory, review material, credentials, or private sources.
+- Polaris integration shall be described as execution integrity, not workload
+  confidentiality.
+
+## Factory authority
+
+- Every factory worker shall declare one role contract from `openvang/factory.py`.
+- No role may automatically access a wallet, submit an on-chain transaction,
+  change emissions, vote in governance, mutate GitHub, or publish.
+- Owner-level effects require a separate external approval and signing boundary;
+  a factory `ActionIntent` is non-executable by design.
+- Role-private memory, including private maintainer-review material, shall never
+  cross role boundaries or enter public, benchmark, or TEE artifacts.
+- Security QA is defensive and isolated; it may propose containment but may not
+  perform an offensive or production mutation.
+
+## Quality gates
+
+- Changes under `agent/`, `benchmark/`, `openvang/`, or `vanguarstew_runtime/`
+  shall include matching tests.
+- `ruff check .` and the relevant offline test suite shall pass before release.
+- Runtime defaults shall remain dry-run, loopback-only, private, and without a
+  GitHub write path.
@@ -7,9 +7,6 @@ All notable changes to this project are documented here. The format is based on
 ## [Unreleased]
 
 ### Added
-- CI contribution policy now auto-closes unapproved contributor PRs that touch the benchmark,
-  operational scripts, documentation, blog posts, or Markdown files. Maintainer-authored changes
-  and changes linked to a pre-approved open issue remain allowed (#2099).
 - Repo-set tooling: **freeze-window value validation** (`min_history >= 1`, non-empty
   `after`/`before`) and `scripts/validate_repo_set.py` CLI to check a repo-set JSON before
   replay (#325).
@@ -234,9 +231,6 @@ All notable changes to this project are documented here. The format is based on
 - M2: the pairwise judge now evaluates the **decision process** — the agent's inferred
   maintainer philosophy and reasoning are passed to the judge and weighed alongside
   trajectory/direction match, so when two plans point the same way the sounder reasoning wins.
-- Trustable contribution pipeline: a published review/scoring rubric (`REVIEW.md`), a
-  PR-integrity check (issue reference, no AI-attribution, non-trivial diff, tests-with-code,
-  per-author PR limit), `CODEOWNERS` review routing, and a CI coverage floor.
 
 ## [0.1.0] - 2026-07-02