OpenClaw maintainer automation
ClawSweeper
AI issue cleanup for OpenClaw
ClawSweeper is a conservative maintainer bot for
openclaw/openclaw. It reviews issues and pull requests,
writes one auditable markdown record per open item, and proposes
closure only when the evidence is strong.
Dashboard numbers were read from the ClawSweeper README on April 25, 2026 at 11:21 UTC.
Project summary
What ClawSweeper actually does
The project is not a general-purpose GitHub bot. It is a focused cleanup system for OpenClaw's own repository, where thousands of issues and PRs can become difficult to triage by hand.
ClawSweeper reads open items, checks them against the current
main branch, asks Codex to make a structured decision,
and stores the outcome in items/<number>.md. The
stored record includes the issue snapshot hash, review status,
confidence, evidence, risks, and any proposed close comment.
The normal mode is proposal-only. Applying closures is a separate path that refetches the GitHub context and skips anything that changed after review.
ClawSweeper guide
Why this project needs more than a README summary
The ClawSweeper README is compact, but the project itself describes a careful maintainer workflow. For SEO and for readers arriving from search, clawsweeper.click explains the operational meaning behind the dashboard numbers, the review cadence, the allowed close policy, and the apply workflow that turns proposals into GitHub actions.
Dashboard signals
The dashboard is not decorative. It reports open OpenClaw issues, open pull requests, reviewed files, proposed closes, archived closed files, failed or stale reviews, and cadence coverage. Those metrics tell maintainers whether ClawSweeper is keeping up with the live backlog or falling behind.
Review cadence
Recent activity is reviewed more aggressively. The README explains that pull requests, items with activity since the last snapshot, and issues younger than 30 days are reviewed daily. Older inactive issues are reviewed weekly, so stale work is still checked without crowding out active development.
Folder reconciliation
ClawSweeper keeps items/ focused on open work and
closed/ focused on archived records. Reconciliation
moves externally closed items into the archive and moves reopened
records back into active tracking as stale, so the next planner can
review them again.
Workflow
How the weekly ClawSweeper review works
The GitHub Actions workflow can run on a schedule or manually. Its default sweep is built around planning, parallel review shards, artifact publishing, and cautious apply runs.
Plan candidates
The planner scans OpenClaw items, prioritizes due activity, and emits explicit item-number batches. Defaults include 40 shards and 5 items per worker.
Review read-only
Each shard checks out openclaw/openclaw at
main, runs Codex with a 10-minute per-item timeout,
and verifies the checkout stays unchanged.
Publish records
Review artifacts are merged into the repository, folder
reconciliation updates items/ and closed/,
and the README dashboard is refreshed.
Apply unchanged proposals
Apply mode reads existing proposed closes, recomputes the snapshot hash, skips changed or maintainer-authored items, and commits checkpoints while respecting GitHub throttling.
ClawSweeper on X
Public posts about ClawSweeper
These six X posts are embedded as official post cards. If X blocks third-party embeds in a browser, each card still links to the original post.
Guardrails
Allowed close reasons are deliberately narrow
Already implemented on main.
Cannot reproduce on current main.
Belongs on ClawHub as a skill or plugin, not in core.
Too incoherent to be actionable.
Stale issue older than 60 days with insufficient data to verify the bug.
Safety model
Built for traceable maintainer cleanup
Codex runs without GitHub write tokens during review.
The OpenClaw checkout is made read-only in CI and checked before and after each review.
Maintainer-authored items from owners, members, and collaborators are excluded from automated closes.
Apply mode defaults to issue cleanup; PR closure is not the default path.
GitHub secondary rate limits trigger long retry backoff and dashboard heartbeat updates.
Audit trail
Every decision becomes a markdown record
The repository's items/ files show why an item was kept
open or proposed for closure. They are structured enough for machines
and readable enough for maintainers.
number: 10005
type: issue
reviewed_at: 2026-04-24T01:35:53.287Z
review_mode: propose
review_status: complete
local_checkout_access: verified
item_snapshot_hash: 7fae9a18...
decision: keep_open
close_reason: none
confidence: high
action_taken: kept_openOperations
How ClawSweeper turns reviews into maintainable history
ClawSweeper is useful because the review output is durable. A maintainer can inspect a record, see the item number, item type, original URL, author, labels, review time, main branch SHA, latest release at review time, confidence level, evidence list, risk notes, and action taken. That structure makes the project easier to audit than a bot that only leaves a short GitHub comment.
Apply checkpoints
Apply mode works in checkpoints. The README describes checkpoint commits, fresh-close limits, progress logging, and a close delay that helps avoid GitHub secondary write throttling. If throttling occurs, ClawSweeper posts a best-effort dashboard heartbeat before retrying.
Local run requirements
Running ClawSweeper locally requires Node 24. The documented local flow builds the TypeScript project, plans candidates, reviews against an OpenClaw checkout, applies artifacts, and can run reconciliation in dry-run mode before maintainers trust any folder movement.
Secrets and responsibility
The GitHub Actions setup separates OpenAI authentication from the GitHub token used to comment and close OpenClaw items. The workflow fails review shards instead of writing fallback markdown when Codex authentication or output fails, which keeps uncertain results out of the record set.
Community context
Notes, tweets, and maintainer commentary
This section is ready for public posts you want to add later. Paste concise quotes, tweet summaries, or maintainer notes here as the ClawSweeper story develops.
"ClawSweeper turns backlog cleanup into auditable proposals instead of silent automation."
"The important design choice is not the AI review; it is the snapshot check before anything closes."
ClawSweeper FAQ
Questions people search for
What is ClawSweeper?
ClawSweeper is a conservative OpenClaw maintainer bot that reviews open GitHub issues and PRs, writes evidence records, and proposes closures only for narrow, high-confidence cases.
Does ClawSweeper close issues automatically?
Not in the normal workflow. The regular review path is proposal-only. A later apply-existing run can close items, but it first verifies that the stored snapshot still matches the current GitHub item.
What technology does ClawSweeper use?
The repository is a private-package TypeScript project for Node 24.
It builds with tsgo, runs through GitHub Actions, and
invokes Codex with gpt-5.4 in the documented workflow.
Why is the keyword clawsweeper important?
ClawSweeper is both the project name and the likely search term
for people trying to understand the openclaw/clawsweeper
GitHub repository, its dashboard, and its issue cleanup model.