[Feature]: verified OSS contribution loop for prioritized, non-duplicate agent work

Problem or Motivation

Agent Canvas can execute repository work, while OpenHands already requires a linked ready-for-dev issue with concrete desired behavior and acceptance criteria before a PR is eligible. The missing operator workflow is between a ready issue and a maintainable agent contribution:

• several Agent Canvas runs can select or attempt the same issue;
• a run can pass focused tests without proving the requested user behavior;
• evidence can become detached from the final PR head after repair or rebase;
• the implementing run can appear to validate its own work;
• maintainers still have to reconstruct trajectory, environment, cost, and limitations from several surfaces.

This makes it hard to run OpenHands against a project priority queue without creating duplicate work or expensive review reconstruction.

Desired Behavior

As a maintainer, I can enable an Agent Canvas contribution workflow that begins only from an issue carrying ready-for-dev:

1. Canvas lists repository-selected ready-for-dev issues in maintainer priority order.
2. Starting a run creates an expiring claim tied to the issue, base SHA, actor, and Canvas run.
3. A competing active claim or PR is visible before another run starts.
4. The run works in an isolated branch/worktree and records its trajectory normally.
5. The final result emits a compact contribution receipt bound to the exact PR head:
   • issue and acceptance criteria;
   • base/head SHAs and changed files;
   • pre-change reproduction when applicable;
   • focused tests plus the closest build or Agent Canvas user path;
   • trajectory/run ID, environment, model usage, and known limitations.
6. A separate review run can inspect the issue, trajectory, receipt, diff, and exact head without writing to that branch.
7. OpenHands reports ready for maintainer review, changes required, duplicate, or blocked; it never awards merge authority.

The existing issue, ready-for-dev automation, PR, and branch rules remain canonical. This is coordination and evidence around Agent Canvas, not another tracker.

Acceptance Criteria

• Only issues carrying ready-for-dev are offered for autonomous implementation.
• A claim records issue, base SHA, actor/run ID, scope, and expiry.
• Active claims and linked PRs are shown before launch; expired claims stop blocking selection.
• Each run uses isolated mutable source state.
• The contribution receipt is regenerated or invalidated whenever the PR head changes.
• Receipt evidence maps directly to the issue's acceptance criteria.
• Evidence distinguishes focused tests, CI, mock/simulated behavior, and a real Agent Canvas path.
• An independent read-only review run verifies the exact head.
• Existing GitHub review and merge rules remain authoritative.
• A pilot demonstrates one successful issue and one duplicate/stale claim.

Alternatives Considered

• Select issues from labels only: ready-for-dev proves readiness, not that nobody else is already implementing it.
• Rely on the PR test section: it does not automatically bind trajectory and runtime evidence to the current head.
• Let each run post free-form logs: expensive for maintainers to compare and easy to make stale.
• Create a separate OpenHands backlog: rejected; the repository's issues and readiness gate should remain the source of truth.

No files or tests are named in the issue. Start by locating the existing ready-for-dev automation and Agent Canvas run or branch entry points, then map claim, receipt, and independent review requirements to them and verify completion against every acceptance criterion, including the pilot.