Target validation hard-codes pnpm + check:changed; bun-based target repos (e.g. openclaw/clawhub) always no-op with validation_script_missing

[momothemage]

Target validation hard-codes pnpm + check:changed; bun-based target repos (e.g. openclaw/clawhub) always no-op with validation_script_missing

Summary

ClawSweeper's target-validation pipeline is built around a single assumption — the target repo is an OpenClaw pnpm monorepo that exposes a check:changed script. Any target repo that uses a different toolchain (in our case openclaw/clawhub, which is bun-based) cannot get past preflight: every automerge/autofix attempt ends as a no-op with

Executor outcome: validation_script_missing: required pnpm check:changed is
unavailable in target checkout.

even though the PR itself is perfectly mergeable.

Repro

1. Open a PR against openclaw/clawhub and opt it into ClawSweeper automerge/autofix.
2. Worker runs preflightTargetValidationPlan against the clawhub checkout.
3. clawhub/package.json does not define a check:changed script (it ships check, check:peers, check:secrets only) and uses bun (bun.lock, no pnpm-lock.yaml, packageManager is not pnpm@…).
4. Worker exits as no-op with the message above. No push, rebase, replacement PR, merge, or re-review happens.

Root cause

Three places in src/repair/target-validation.ts hard-wire the pnpm + OpenClaw shape:

1. prepareTargetToolchain (line ~57) throws unsupported target package manager: ${packageManager} whenever package.json#packageManager doesn't start with pnpm@. Bun targets are rejected up front.

2. requiredValidationCommands (~line 211) unions the fixArtifact's validation_commands with the OpenClaw changed-gate. The gate itself is guarded by requiresOpenClawChangedGate (only fires when targetRepo === "openclaw/openclaw" and the repo defines check:changed), but the prompts in prompts/repair/autonomous.md and docs/repair/automerge-flow.md describe pnpm check:changed as the default local gate, so the LLM-generated fixArtifact.validation_commands carries pnpm check:changed for every OpenClaw-org repo, not just openclaw/openclaw.

3. preflightTargetValidationPlan (~line 197-225) then resolves each command to its required npm script via packageScriptRequirement, compares it against the keys of the target's package.json#scripts, and emits

   {
     status: "blocked",
     code: "validation_script_missing",
     required: missing.command,        // "pnpm check:changed"
     missing_script: missing.name,     // "check:changed"
     reason: `validation_script_missing: required ${missing.command} is unavailable in target checkout`,
   }
   

   execute-fix-artifact.ts treats this as a non-retryable terminal outcome (see the /no merge base|validation_script_missing/i short-circuit around line 2681), so the worker no-ops.

The net effect: any PR targeting a non-pnpm / non-check:changed repo is unreachable by automerge, even when the PR has zero validation problems.

Why config alone is not enough

config/target-repositories.json doesn't currently override the validation command set or the package-manager check — prepareTargetToolchain and preflightTargetValidationPlan ignore it for these decisions. So even if we add an entry for openclaw/clawhub there, the pnpm/check:changed assumptions still fire.

Proposed fix (config-driven, smallest blast radius)

Make the validation pipeline read three things per target repo from config/target-repositories.json:

• packageManager: "pnpm" | "bun" | "npm" — gate prepareTargetToolchain (corepack vs. bun install) on this instead of a hard pnpm@ prefix check.
• validationCommands: string[] — used as the base set inside requiredValidationCommands; default still adds pnpm check:changed for openclaw/openclaw, but other repos opt in explicitly (e.g. clawhub → ["bun run check"]).
• changedGate: { command, requiredScript } | null — replaces the hard-coded pnpm check:changed / check:changed pair in requiresOpenClawChangedGate and resolveAllowedValidationCommands. Repos with null skip the gate normalization entirely.

Then:

• prepareTargetToolchain: branch on packageManager. For bun, run bun install --frozen-lockfile (with --no-frozen-lockfile fallback on lockfile drift), no corepack.
• requiredValidationCommands: add the configured changedGate.command instead of literal "pnpm check:changed".
• preflightTargetValidationPlan / packageScriptRequirement: keep the existing logic, just consult the per-repo config when deciding which script name a given command implies.
• validationFallbackCommands / shouldRetryValidationCommand: parameterize the parts[0] !== "pnpm" || parts[1] !== "check:changed" check on the configured changed-gate so bun targets also get the merge-base recovery and the single transient retry.

Tests in test/repair/target-validation.test.ts already drive the pnpm path end-to-end via packageFixture — adding a sibling bunPackageFixture plus a clawhub-shaped repo entry should give us the new path coverage cheaply.

Workaround until fixed

Maintainers have to merge clawhub PRs manually; ClawSweeper automerge cannot reach them.

References

• src/repair/target-validation.ts — prepareTargetToolchain (≈ L43-L100), requiredValidationCommands (≈ L211-L219), preflightTargetValidationPlan (≈ L168-L225), requiresOpenClawChangedGate (≈ L455).
• src/repair/execute-fix-artifact.ts — terminal-outcome detector at if (/no merge base|validation_script_missing/i.test(message)) return false;
• prompts/repair/autonomous.md L69, docs/repair/automerge-flow.md L108 — prompt-side source of the pnpm check:changed default that bleeds into non-openclaw target repos.

Suggested labels

area: repair, area: target-validation, bug, multi-repo

[clawsweeper]

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed June 1, 2026, 9:57 PM ET / 01:57 UTC.

Summary
Keep open: current main still has the reported repair target-validation gap. The source path rejects non-pnpm target package managers, the preflight path still blocks on an artifact-provided pnpm check:changed when the target lacks that script, and the target repository profile/config schema has no validation/toolchain metadata to override it.

Reproducibility: yes. at source level: a bun packageManager reaches the non-pnpm throw in prepareTargetToolchain, and a pnpm check:changed validation command reaches validation_script_missing when the target scripts omit check:changed. I did not run a live repair/automerge attempt.

Next step
A focused repair PR can extend target repository metadata and validation handling without needing a product decision.

Review details

Best possible solution:

Add target repository validation metadata and make the repair executor use it for package-manager setup, base validation commands, script requirements, changed-gate normalization, and fallback/retry behavior while preserving the existing OpenClaw-only safety gate.

Do we have a high-confidence way to reproduce the issue?

Yes, at source level: a bun packageManager reaches the non-pnpm throw in prepareTargetToolchain, and a pnpm check:changed validation command reaches validation_script_missing when the target scripts omit check:changed. I did not run a live repair/automerge attempt.

Is this the best way to solve the issue?

Yes, a target-profile/config-driven fix is the narrow maintainable direction because the OpenClaw pnpm check:changed gate should stay repo-specific instead of leaking into every target checkout.

AGENTS.md: found and applied where relevant.

Codex review notes: model gpt-5.5, reasoning high; reviewed against a07fc1f94275.

Label changes

Label changes:

• add P2: The issue blocks repair/automerge validation for a supported non-pnpm target repository but does not indicate data loss, security impact, or a core runtime outage.
• add impact:other: The affected behavior is maintainer-visible automation reachability for target repositories outside the more specific impact taxonomy.
• add issue-rating: 🦞 diamond lobster: Current issue advisory state selects this label.
• add clawsweeper:source-repro: Current issue advisory state selects this label.
• add clawsweeper:queueable-fix: Current issue advisory state selects this label.
• add clawsweeper:fix-shape-clear: Current issue advisory state selects this label.

Label justifications:

• P2: The issue blocks repair/automerge validation for a supported non-pnpm target repository but does not indicate data loss, security impact, or a core runtime outage.
• impact:other: The affected behavior is maintainer-visible automation reachability for target repositories outside the more specific impact taxonomy.

Evidence reviewed

Acceptance criteria:

• pnpm run check

What I checked:

• AGENTS policy read: Read the full target AGENTS.md; its repair-lane and target repository profile guidance applies to this issue. (AGENTS.md:1, a07fc1f94275)
• Non-pnpm target package managers are rejected: prepareTargetToolchain reads package.json#packageManager, defaults to pnpm@10.33.0, and throws when the value does not start with pnpm@. (src/repair/target-validation.ts:54, a07fc1f94275)
• Preflight can still block on missing check:changed: preflightTargetValidationPlan resolves artifact validation commands, maps pnpm check:changed to a required package script, and returns validation_script_missing when that script is absent. (src/repair/target-validation.ts:180, a07fc1f94275)
• Package-script detection is pnpm/npm-only: packageScriptRequirement recognizes npm run <script> and pnpm <script> but ignores bun commands, so bun validation is not represented in the current script requirement model. (src/repair/validation-command-utils.ts:6, a07fc1f94275)
• ClawHub target config has no validation/toolchain override: The openclaw/clawhub config entry defines checkout/display/prompt/close rules only; no package-manager, validation-command, or changed-gate fields are present. (config/target-repositories.json:5, a07fc1f94275)
• Repository profile schema cannot currently carry the requested data: RepositoryProfile and ConfiguredRepositoryProfile contain display, checkout, URL, prompt, and close-rule fields, while currentTargetValidationOptions is built from job/environment state rather than target config. (src/repository-profiles.ts:18, a07fc1f94275)

Likely related people:

• Takhoffman: Blame and log history tie the current repair target-validation implementation, prompt guidance, and ClawHub target config to commit 9b0b0b39173365a6d82cb4dbc59b21de3fd54270. (role: introduced behavior; confidence: high; commits: 9b0b0b391733; files: src/repair/target-validation.ts, src/repair/fix-prompt-builder.ts, src/repair/commit-finding-intake.ts)

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.