Summary

Adds a phase 1 proof-followup lane for pull requests that are stuck on triage: needs-real-behavior-proof.

This is intentionally v1 nudges only:

• no dashboard work
• no auto-close behavior
• no label churn for cooldowns
• no repair or merge behavior
• dry-run by default for manual runs
• scheduled operation is present but off by default behind repository variables

What This Does

Adds a new proof-nudges ClawSweeper command that finds open PRs which still have the live triage: needs-real-behavior-proof label and are old enough to deserve a polite reminder.

The nudge asks for real behavior evidence, gives examples of acceptable proof, and points contributors toward maintainer re-review after they update the PR.

Safety Rules

The nudge lane skips PRs when any of these apply:

• PR is closed, locked, stale relative to the report head, or missing the live proof-needed label
• proof is already supplied, sufficient, overridden, or the proof policy is not applicable
• author has commented recently
• PR body or head commit was updated recently
• ClawSweeper reviewed the PR recently
• a same-head ClawSweeper cooldown marker already exists in a prior bot comment
• PR is authored by a maintainer or bot
• PR appears to be security-sensitive, release-related, or otherwise protected

Cooldown state is kept in a hidden marker in ClawSweeper's own comment body rather than by adding/removing labels.

Workflow

Adds a proof-nudge lane under the existing sweep workflow.

Manual use:

• set proof_nudges=true when dispatching the workflow
• mode defaults to dry-run because proof_nudges_execute=false
• set proof_nudges_execute=true only when comments should actually be posted
• proof_nudges_limit controls the batch size, default 10

Scheduled use is also wired, but deliberately disabled by default:

• the workflow includes a daily cron at 41 9 * * *
• scheduled runs do nothing unless CLAWSWEEPER_PROOF_NUDGES_SCHEDULED=1 is set
• scheduled runs remain dry-run only unless CLAWSWEEPER_PROOF_NUDGES_EXECUTE=1 is also set
• optional vars can tune CLAWSWEEPER_PROOF_NUDGES_TARGET_REPO, CLAWSWEEPER_PROOF_NUDGES_LIMIT, CLAWSWEEPER_PROOF_NUDGES_MIN_AGE_DAYS, and CLAWSWEEPER_PROOF_NUDGES_COOLDOWN_DAYS

Suggested rollout is manual dry-run first, then scheduled dry-run, then scheduled execute only once maintainers are happy with the reports.

Docs

Adds docs/proof-nudges.md and links it from the README. The docs include the manual command, workflow inputs, default-off scheduled vars, and rollout path.

Validation

• pnpm run build:all
• pnpm run lint:src
• pnpm run lint:scripts
• pnpm exec oxfmt --check README.md src\clawsweeper.ts test\clawsweeper.test.ts package.json docs\proof-nudges.md .github\workflows\sweep.yml
• node --test --test-name-pattern "proof nudge" test\clawsweeper.test.ts
• Codex review completed cleanly with no accepted actionable findings