openclaw/clawsweeper
openclaw/clawsweeper
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 2 Wiki Activity

fix: dispatch repair worker workflow

Browse Source
This commit is contained in:
Peter Steinberger 2026-04-29 13:58:00 +01:00
parent 692bdd98f4
commit 0ef767cf5e
No known key found for this signature in database
13 changed files with 50 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
docs/repair/README.md
View File

@ -226,7 +226,7 @@ For a maintainer-facing architecture map of the automation lanes, see
[`docs/INTERNAL_FEATURES.md`](docs/INTERNAL_FEATURES.md).
For the ClawSweeper feedback loop that updates existing generated PRs, see
[`docs/auto-update-prs.md`](docs/auto-update-prs.md).
[`docs/repair/auto-update-prs.md`](auto-update-prs.md).
That loop is marker-driven. ClawSweeper comments use hidden
`clawsweeper-verdict:*` markers, and only actionable PR feedback includes
@ -312,7 +312,7 @@ Supported commands:
```
`status` and `explain` post a short status reply. `fix ci`, `address review`,
and `rebase` dispatch the normal `cluster-worker.yml` repair path, but only for
and `rebase` dispatch the normal `repair-cluster-worker.yml` repair path, but only for
existing ClawSweeper PRs identified by the `clawsweeper/*` branch.
`automerge` opts an open PR into the bounded review/fix/merge loop. `approve`
is maintainer-only exact-head approval after a human-review pause; it clears

14
docs/repair/auto-update-prs.md
View File

@ -21,7 +21,7 @@ The loop is intentionally small:
to review that PR head.
3. The comment router sees trusted ClawSweeper feedback.
4. ClawSweeper dispatches the existing or adopted job through
`cluster-worker.yml`.
`repair-cluster-worker.yml`.
5. The repair worker pushes another commit to the source branch if it finds a
safe, narrow fix, or opens a credited replacement when the source branch
cannot be safely updated.
@ -177,7 +177,7 @@ five automatic ClawSweeper-triggered repair iterations. The per-PR cap is total
across all head SHAs and stops the automatic review/repair loop even when every
iteration produces a new commit.
Runs for the same job path and mode share the `cluster-worker.yml` concurrency
Runs for the same job path and mode share the `repair-cluster-worker.yml` concurrency
group, so repeated dispatches queue instead of racing the same branch.
ClawSweeper edits one durable review comment in place. The router keys its
@ -227,13 +227,13 @@ as:
Workflow:
- `.github/workflows/comment-router.yml`
- `.github/workflows/repair-comment-router.yml`
Scripts:
- `scripts/comment-router.ts`
- `scripts/comment-router-core.ts`
- `scripts/comment-router-utils.ts`
- `src/repair/comment-router.ts`
- `src/repair/comment-router-core.ts`
- `src/repair/comment-router-utils.ts`
Durable state:
@ -255,7 +255,7 @@ Syntax and workflow checks:
```bash
pnpm run check
actionlint .github/workflows/comment-router.yml
actionlint .github/workflows/repair-comment-router.yml
```
Dry-run the router against live recent comments:

18
docs/repair/internal-features.md
View File

@ -106,7 +106,7 @@ replacement PR. Direct mutation still happens outside Codex.
## Cloud Worker Flow
Workflow: `.github/workflows/cluster-worker.yml`
Workflow: `.github/workflows/repair-cluster-worker.yml`
The cluster worker has two jobs:
@ -285,7 +285,7 @@ The finalizer scans open ClawSweeper PRs in the target repo. It finds PRs by the
- security hold
When `--dispatch-repairs --execute` is enabled, it dispatches the existing
cluster job back through `cluster-worker.yml` instead of creating another PR.
cluster job back through `repair-cluster-worker.yml` instead of creating another PR.
The idempotency key includes target repo, PR number, and head SHA, so the same
PR/head is not repeatedly repaired unless `--allow-repeat` is used.
@ -295,8 +295,8 @@ clearly transient jobs, and pass branch-caused failures into the repair prompt.
## Self-Heal Failed ClawSweeper Runs
Workflow: `.github/workflows/self-heal.yml`
Script: `scripts/self-heal-failed-runs.ts`
Workflow: `.github/workflows/repair-self-heal.yml`
Script: `src/repair/self-heal-failed-runs.ts`
Self-heal retries failed ClawSweeper cluster-worker runs. It reads published
`results/runs/*.json`, selects the latest failed run per source job, skips jobs
@ -309,11 +309,11 @@ finalizer/comment command repair path.
## Maintainer Comment Routing
Workflow: `.github/workflows/comment-router.yml`
Workflow: `.github/workflows/repair-comment-router.yml`
Scripts:
- `scripts/comment-router.ts`
- `scripts/comment-router-core.ts`
- `src/repair/comment-router.ts`
- `src/repair/comment-router-core.ts`
Comment routing scans recent target-repo issue/PR comments and accepts only
maintainer-authored commands. Default allowed GitHub `author_association`
@ -326,7 +326,7 @@ values:
Contributor comments are ignored without a reply.
The generated-PR auto-update design is documented in
[`docs/auto-update-prs.md`](auto-update-prs.md). That lane lets trusted
[`docs/repair/auto-update-prs.md`](auto-update-prs.md). That lane lets trusted
ClawSweeper comments dispatch a repair run for an existing ClawSweeper PR or a
PR explicitly opted into `clawsweeper:automerge` without allowing arbitrary
comment authors to trigger work.
@ -372,7 +372,7 @@ Behavior:
Repair commands apply to existing ClawSweeper PRs and PRs opted into
`clawsweeper:automerge`. The router finds ClawSweeper PRs by the
`clawsweeper/*` branch, resolves or creates the backing job, posts one
idempotent response marker, and dispatches `cluster-worker.yml`.
idempotent response marker, and dispatches `repair-cluster-worker.yml`.
Trusted ClawSweeper comments become `clawsweeper_auto_repair`. Preferred
comments use hidden `clawsweeper-verdict:*` markers and include

4
docs/repair/operations.md
View File

@ -5,7 +5,7 @@ commands, finalizers, self-heal, gates, and ledgers, see
[`docs/INTERNAL_FEATURES.md`](INTERNAL_FEATURES.md).
For the trusted ClawSweeper-to-ClawSweeper PR repair loop, see
[`docs/auto-update-prs.md`](auto-update-prs.md).
[`docs/repair/auto-update-prs.md`](auto-update-prs.md).
For commit-review findings, ClawSweeper dispatches
`clawsweeper_commit_finding` to this repository. ClawSweeper fetches the latest
@ -194,7 +194,7 @@ Repair commands apply to existing ClawSweeper PRs and to PRs opted into
`clawsweeper/*` branch prefix. Opted-in non-ClawSweeper PRs get an adopted job
at `jobs/<owner>/inbox/automerge-<owner>-<repo>-<pr>.md`.
The router posts one idempotent reply with a hidden marker and dispatches the
normal `cluster-worker.yml` repair path. It records processed comment versions
normal `repair-cluster-worker.yml` repair path. It records processed comment versions
in `results/comment-router.json`. For durable ClawSweeper comments,
idempotency is per comment id plus GitHub `updated_at`, and response markers
include the target PR head SHA. That lets edited ClawSweeper comments wake

11
src/repair/config.ts
View File

@ -2,7 +2,12 @@ import type { JsonValue, LooseRecord } from "./json-types.js";
import { DEFAULT_ALLOWED_REPOSITORY_PERMISSIONS } from "./comment-router-core.js";
import { currentProjectRepo, readMaxLiveWorkers } from "./lib.js";
import { assertRepo, commaSet, positiveInteger } from "./comment-router-utils.js";
import { DEFAULT_HEAD_PREFIX, DEFAULT_TARGET_REPO } from "./constants.js";
import {
DEFAULT_HEAD_PREFIX,
DEFAULT_TARGET_REPO,
REPAIR_CLUSTER_WORKFLOW,
SWEEP_WORKFLOW,
} from "./constants.js";
export { DEFAULT_HEAD_PREFIX, DEFAULT_TARGET_REPO } from "./constants.js";
const DEFAULT_ALLOWED_ASSOCIATIONS = ["OWNER", "MEMBER", "COLLABORATOR"];
@ -44,7 +49,7 @@ export function readCommentRouterConfig(args: LooseRecord): CommentRouterConfig
);
const workflow = stringSetting(
args.workflow ?? process.env.CLAWSWEEPER_COMMENT_WORKFLOW,
"cluster-worker.yml",
REPAIR_CLUSTER_WORKFLOW,
);
const reviewRepo = stringSetting(
args["review-repo"] ?? process.env.CLAWSWEEPER_REVIEW_REPO,
@ -52,7 +57,7 @@ export function readCommentRouterConfig(args: LooseRecord): CommentRouterConfig
);
const reviewWorkflow = stringSetting(
args["review-workflow"] ?? process.env.CLAWSWEEPER_REVIEW_WORKFLOW,
"sweep.yml",
SWEEP_WORKFLOW,
);
const runner = stringSetting(
args.runner ?? process.env.CLAWSWEEPER_WORKER_RUNNER,

3
src/repair/constants.ts
View File

@ -2,6 +2,9 @@ export const DEFAULT_TARGET_REPO = "openclaw/openclaw";
export const DEFAULT_HEAD_PREFIX = "clawsweeper/";
export const DEFAULT_LABEL = "clawsweeper";
export const REPAIR_CLUSTER_WORKFLOW = "repair-cluster-worker.yml";
export const SWEEP_WORKFLOW = "sweep.yml";
export const CLAWSWEEPER_LABEL = "clawsweeper";
export const CLAWSWEEPER_LABEL_COLOR = "F97316";
export const CLAWSWEEPER_LABEL_DESCRIPTION = "Tracked by ClawSweeper automation";

3
src/repair/dispatch-jobs.ts
View File

@ -15,6 +15,7 @@ import {
waitForLiveWorkerCapacity,
} from "./lib.js";
import { sleepMs } from "./timing.js";
import { REPAIR_CLUSTER_WORKFLOW } from "./constants.js";
const args = parseArgs(process.argv.slice(2));
const defaultRunner = process.env.CLAWSWEEPER_WORKER_RUNNER ?? "blacksmith-4vcpu-ubuntu-2404";
@ -23,7 +24,7 @@ const defaultExecutionRunner =
const mode = args.mode ?? "plan";
const runner = args.runner ?? defaultRunner;
const executionRunner = args["execution-runner"] ?? args.execution_runner ?? defaultExecutionRunner;
const workflow = args.workflow ?? "cluster-worker.yml";
const workflow = args.workflow ?? REPAIR_CLUSTER_WORKFLOW;
const repo = String(args.repo ?? currentProjectRepo());
const model = String(args.model ?? process.env.CLAWSWEEPER_MODEL ?? "gpt-5.5");
const maxLiveWorkers = readMaxLiveWorkers(args);

4
src/repair/finalize-open-prs.ts
View File

@ -14,7 +14,7 @@ import {
} from "./lib.js";
import { ghJson, ghText } from "./github-cli.js";
import { sleepMs } from "./timing.js";
import { DEFAULT_TARGET_REPO, REVIEW_BOTS } from "./constants.js";
import { DEFAULT_TARGET_REPO, REPAIR_CLUSTER_WORKFLOW, REVIEW_BOTS } from "./constants.js";
import { numberEnv } from "./env-utils.js";
import { compactText, escapeRegExp } from "./text-utils.js";
@ -35,7 +35,7 @@ const writeReport = Boolean(args["write-report"]);
const execute = Boolean(args.execute);
const dispatchRepairs = Boolean(args["dispatch-repairs"] || args.dispatch || execute);
const workflow = String(
args.workflow ?? process.env.CLAWSWEEPER_FINALIZER_WORKFLOW ?? "cluster-worker.yml",
args.workflow ?? process.env.CLAWSWEEPER_FINALIZER_WORKFLOW ?? REPAIR_CLUSTER_WORKFLOW,
);
const runner = String(
args.runner ?? process.env.CLAWSWEEPER_WORKER_RUNNER ?? "blacksmith-4vcpu-ubuntu-2404",

9
src/repair/live-worker-capacity.ts
View File

@ -1,5 +1,6 @@
import { ghJson } from "./github-cli.js";
import type { JsonValue, LooseRecord } from "./json-types.js";
import { REPAIR_CLUSTER_WORKFLOW } from "./constants.js";
import { currentProjectRepo } from "./project-repo.js";
import { sleepMs } from "./timing.js";
@ -20,7 +21,7 @@ export function readMaxLiveWorkers(args: LooseRecord = {}) {
export function liveWorkerCapacity({
repo = currentProjectRepo(),
workflow = "cluster-worker.yml",
workflow = REPAIR_CLUSTER_WORKFLOW,
requested = 1,
maxLiveWorkers = DEFAULT_MAX_LIVE_WORKERS,
}: LooseRecord = {}) {
@ -61,7 +62,7 @@ export function waitForLiveWorkerCapacity(options: LooseRecord = {}) {
);
if (requestedCount > max) {
throw new Error(
`refusing dispatch: requested ${requestedCount} ${options.workflow ?? "cluster-worker.yml"} workers exceeds max-live-workers=${max}`,
`refusing dispatch: requested ${requestedCount} ${options.workflow ?? REPAIR_CLUSTER_WORKFLOW} workers exceeds max-live-workers=${max}`,
);
}
const pollMs = readPositiveInteger(
@ -91,13 +92,13 @@ export function waitForLiveWorkerCapacity(options: LooseRecord = {}) {
}
throw new Error(
`timed out waiting for ${options.workflow ?? "cluster-worker.yml"} capacity: ${latest?.active ?? "unknown"} active + ${requestedCount} requested exceeds max-live-workers=${max}`,
`timed out waiting for ${options.workflow ?? REPAIR_CLUSTER_WORKFLOW} capacity: ${latest?.active ?? "unknown"} active + ${requestedCount} requested exceeds max-live-workers=${max}`,
);
}
export function listActiveWorkflowRuns({
repo = currentProjectRepo(),
workflow = "cluster-worker.yml",
workflow = REPAIR_CLUSTER_WORKFLOW,
}: LooseRecord = {}) {
const runs: LooseRecord[] = [];
for (const status of ACTIVE_WORKFLOW_STATUSES) {

3
src/repair/requeue-job.ts
View File

@ -16,9 +16,10 @@ import {
} from "./lib.js";
import { ghJson, ghText } from "./github-cli.js";
import { sleepMs } from "./timing.js";
import { REPAIR_CLUSTER_WORKFLOW } from "./constants.js";
const DEFAULT_REPO = currentProjectRepo();
const DEFAULT_WORKFLOW = "cluster-worker.yml";
const DEFAULT_WORKFLOW = REPAIR_CLUSTER_WORKFLOW;
const DEFAULT_RUNNER = process.env.CLAWSWEEPER_WORKER_RUNNER ?? "blacksmith-4vcpu-ubuntu-2404";
const DEFAULT_EXECUTION_RUNNER =
process.env.CLAWSWEEPER_EXECUTION_RUNNER ?? "blacksmith-16vcpu-ubuntu-2404";

3
src/repair/self-heal-failed-runs.ts
View File

@ -15,9 +15,10 @@ import {
} from "./lib.js";
import { ghJson, ghText } from "./github-cli.js";
import { sleepMs } from "./timing.js";
import { REPAIR_CLUSTER_WORKFLOW } from "./constants.js";
const DEFAULT_REPO = currentProjectRepo();
const DEFAULT_WORKFLOW = "cluster-worker.yml";
const DEFAULT_WORKFLOW = REPAIR_CLUSTER_WORKFLOW;
const DEFAULT_RUNNER = process.env.CLAWSWEEPER_WORKER_RUNNER ?? "blacksmith-4vcpu-ubuntu-2404";
const DEFAULT_EXECUTION_RUNNER =
process.env.CLAWSWEEPER_EXECUTION_RUNNER ?? "blacksmith-16vcpu-ubuntu-2404";

3
src/repair/sweep-openclaw-jobs.ts
View File

@ -4,6 +4,7 @@ import fs from "node:fs";
import path from "node:path";
import { hasSecuritySignalText, parseArgs, parseJob, repoRoot, validateJob } from "./lib.js";
import { ghJson } from "./github-cli.js";
import { REPAIR_CLUSTER_WORKFLOW } from "./constants.js";
import { readJsonFileIfExists as readJson } from "./json-file.js";
const args = parseArgs(process.argv.slice(2));
@ -291,7 +292,7 @@ function readActiveClusterRuns() {
"--repo",
repo,
"--workflow",
"cluster-worker.yml",
REPAIR_CLUSTER_WORKFLOW,
"--status",
status,
"--limit",

8
test/repair/comment-router-core.test.mjs
View File

@ -289,7 +289,7 @@ test("renderResponse reports trusted repair dispatches without losing guardrails
target: { head_sha: "def456" },
},
{
workflow: "cluster-worker.yml",
workflow: "repair-cluster-worker.yml",
job_path: "jobs/openclaw/inbox/example.md",
mode: "autonomous",
model: "gpt-5.5",
@ -298,7 +298,7 @@ test("renderResponse reports trusted repair dispatches without losing guardrails
assert.match(body, /Thanks, ClawSweeper/);
assert.match(body, /clawsweeper-command:456:2026-04-29T07:12:31Z:clawsweeper_auto_repair:def456/);
assert.match(body, /cluster-worker\.yml/);
assert.match(body, /repair-cluster-worker\.yml/);
assert.match(body, /safe credited replacement/);
assert.match(body, /narrow fix/);
assert.doesNotMatch(body, /ClawSweeper Repair/i);
@ -359,7 +359,7 @@ test("renderResponse reports automerge repair dispatches", () => {
target: { head_sha: "def457" },
},
{
workflow: "cluster-worker.yml",
workflow: "repair-cluster-worker.yml",
job_path: "jobs/openclaw/inbox/automerge-openclaw-openclaw-74156.md",
mode: "autonomous",
model: "gpt-5.5",
@ -367,7 +367,7 @@ test("renderResponse reports automerge repair dispatches", () => {
);
assert.match(body, /picked up the repair feedback/);
assert.match(body, /cluster-worker\.yml/);
assert.match(body, /repair-cluster-worker\.yml/);
assert.match(body, /automerge-openclaw-openclaw-74156/);
assert.doesNotMatch(body, /did not dispatch/);
});