Codex review: needs real behavior proof before merge. Reviewed June 28, 2026, 11:57 AM ET / 15:57 UTC.

Summary
The branch adds stripLeadingExecDisplayVerb, applies it to exec/bash failure-warning metadata in buildEmbeddedRunPayloads, and adds focused exec-display and payload tests.

PR surface: Source +49, Tests +62. Total +111 across 4 files.

Reproducibility: yes. at source level: current main formats exec failure warnings from lastToolError.meta, and current exec summaries can include run python3 .... I did not establish a live cron/chat/CLI reproduction in this read-only review.

Review metrics: none identified.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #97319
Summary: This PR is a candidate fix for the canonical exec failure-message ambiguity; related PRs target the same issue, but none is merged or safely supersedes this branch yet.

Members:

• canonical: [Bug]: exec tool failure message format makes framework labels indistinguishable from agent-typed commands #97319 - The PR uses closing syntax and targets the exact exec failure-warning ambiguity reported there.
• partial_overlap: fix: clarify exec tool failure warnings #97511 - This open draft PR targets the same issue with a broader formatting approach, but it is still draft and has review work remaining.
• partial_overlap: fix(agents): remove misleading run prefix from exec display for interpreter commands (fixes #97319) #97326 - This open PR targets the same issue through a narrower interpreter-summary change and has not superseded this PR.
• superseded: fix(exec): keep framework verb out of backtick-wrapped failure message (#97319) #97359 - This closed earlier same-author attempt targeted the same issue and was replaced by later work in this branch.
• adjacent_distinct: [Bug]: exec tool treats subprocess nonzero exit codes as tool failures, masking stdout and exit-code semantics #97318 - It came from the same production incident but concerns exec nonzero-exit semantics rather than warning presentation.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge readiness
Overall: 🧂 unranked krab
Proof: 🧂 unranked krab
Patch quality: 🐚 platinum hermit
Result: blocked until stronger real behavior proof is added.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• [P1] Add redacted real cron/chat/CLI output, screenshot, recording, log, or linked artifact showing the corrected after-fix warning; redact private details such as API keys, phone numbers, IP addresses, and non-public endpoints, then update the PR body to trigger re-review or ask a maintainer to comment @clawsweeper re-review.

Proof guidance:

• [P1] Needs stronger real behavior proof before merge: Tests and a one-off tsx simulation are present, but no redacted real cron/chat/CLI capture shows the corrected after-fix warning in a user path. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

Mantis proof suggestion
A real failed exec run through a visible cron/chat/CLI path would materially close the remaining proof gap. A maintainer can ask Mantis to capture proof by posting this exact PR comment:

@openclaw-mantis visual task: verify a failed exec command renders without the framework run label inside the command text.

Risk before merge

• [P1] No redacted real cron/chat/CLI output, screenshot, recording, log, or linked artifact shows the corrected after-fix warning through a real user path yet.

Maintainer options:

1. Decide the mitigation before merge
   Keep the focused fix open until real user-path proof is added, while maintainers compare it with the broader draft formatting PR once that branch is ready.
2. Pause or close
   Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge

• [P1] The remaining blocker is contributor-supplied real behavior proof and maintainer comparison with the broader draft, not a concrete automated code repair.

Security
Cleared: The diff changes TypeScript display formatting and colocated tests only; no dependency, CI, secret, install, or code-execution surface changed.

Real behavior proof — exec failure verb stripping

stripLeadingExecDisplayVerb is covered by unit tests (src/agents/tool-display.test.ts) and the integrated failure-message path is covered by payloads.test.ts + payloads.errors.test.ts. Local run (Node 22.22.1):

$ pnpm test src/agents/tool-display.test.ts src/agents/embedded-agent-runner/run/payloads.test.ts src/agents/embedded-agent-runner/run/payloads.errors.test.ts

 Test Files  3 passed (3)
      Tests  132 passed (132)
   Duration  5.36s
[test] passed 1 Vitest shard in 24.69s

Pre-fix vs post-fix (issue #97319)

The heartbeat-exec payload test was failing on the old wording because the production fix correctly strips the leading framework verb before backtick-wrapping. Updated the test to match the corrected shape:

- title: "show last 20 lines",  // pre-fix expected the verb inside the wrapper
+ title: "last 20 lines",        // post-fix: verb stays outside the backticks

The resulting payload text is now ⚠️ 🛠️ \last 20 lines of ~/.openclaw/workspace/memory/2026-06-04.md` failed: …instead of⚠️ 🛠️ `show last 20 lines of …` failed: …, so the framework's show` label is no longer visually mixed with the user's command subject inside the backtick region.

Coverage of stripLeadingExecDisplayVerb

• Strips known exec display verbs (run, check, view, show, list, …) — tested.
• Passes raw shell commands through unchanged (cd ~/dir && ls, git status, npm install) — tested.
• Single-token and whitespace-only input returned unchanged — tested.

The CI failure that flagged this PR earlier was just the stale test assertion; the production code was already correct.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

@clawsweeper re-review

Real-behavior proof (post-fix) for issue #97319, generated by invoking production stripLeadingExecDisplayVerb from src/agents/tool-display-exec.ts and simulating the exec-failure branch added in src/agents/embedded-agent-runner/run/payloads.ts:551-562 via tsx.

stripLeadingExecDisplayVerb(meta) on production display-meta inputs (the strings summarizeKnownExec emits and stores in lastToolError.meta):

Composed failure-warning subject (mirroring buildEmbeddedRunPayloads exec branch at payloads.ts:551-562):

toolName=exec meta="run python3 /tmp/check.py" -> warning subject: `python3 /tmp/check.py`
toolName=bash meta="run node script ./server.js" -> warning subject: `node script ./server.js`
toolName=exec meta="check git status"           -> warning subject: `git status`

For non-exec tools (isExecLikeToolName(toolName) === false), displayMeta falls back to the original meta unchanged — the strip is gated on the tool family.

The reported case from #97319 (run python3 /path shown as `run python3 /path` in the failure warning, indistinguishable from an agent-typed run … invocation) is fixed: the framework-injected verb run stays outside the backticks.

Local: ./node_modules/.bin/tsx src/agents/exec-verb-proof.mts on Node v22.22.1 (script removed after running — one-off proof harness, not part of the patch).

@clawsweeper re-review

Narrowed the helper in response to CS's P2 finding (commit 6c6181b1).

Change: stripLeadingExecDisplayVerb now only collapses the framework-injected run verb, and only when followed by a known interpreter/binary (node, python, git, npm, openclaw, sed, etc.). The 23 other verbs (check, view, show, install, start, …) are preserved, along with run when followed by a non-binary noun (run tests, run build).

Effect on CS's named examples

Tests (src/agents/tool-display.test.ts): 9 cases covering run+binary collapse, action-verb preservation (install dependencies, start app, run tests, run build, show …, check …, find …), and unchanged fallback for shell-builtins and single tokens. 82 tests pass on Node v22.22.1.

Heartbeat assertion restored (src/agents/embedded-agent-runner/run/payloads.test.ts:476-479): the previously-collapsed last 20 lines is back to show last 20 lines, since show is now preserved.

The reported case from #97319 stays fixed; the broader action-label regression CS flagged is gone.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

P2 follow-up landed in 5ad2ead.

• src/agents/tool-display-exec.ts:331 — renamed the local binaryName (which shadowed the imported binaryName helper) to resolvedBinName.
• src/agents/embedded-agent-runner/run/payloads.errors.test.ts:784 — restored the preserved show verb in the matrix-progress warning expectation so the diagnostic verb stays visible inside the backtick-wrapped failure message.

Proof:

• node_modules/.bin/vitest run src/agents/embedded-agent-runner/run/payloads.errors.test.ts src/agents/tool-display.test.ts — 100/100 pass.
• node scripts/run-oxlint.mjs <both files> — clean.

The earlier P1 real-runtime proof gap (cron/chat/CLI path) remains outstanding and is still blocked on a real user-path capture; happy to add it if a maintainer can share the redacted output format they want.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

P2 consistency follow-up landed in f27953b.

Replaced the binary allowlist with a denylist of semantic run words emitted by the npm/pnpm/yarn subcommand map (tests, build, lint, script). Generic run <binary> patterns now strip consistently — run my-custom-tool --flag collapses to my-custom-tool --flag the same way run python3 /path does — while package-manager script labels keep their run prefix so the failure message preserves script context.

Proof:

• node_modules/.bin/vitest run src/agents/tool-display.test.ts src/agents/embedded-agent-runner/run/payloads.errors.test.ts src/agents/embedded-agent-runner/run/payloads.test.ts — 136/136 pass.
• New coverage: run my-custom-tool --flag value → my-custom-tool --flag value; run ./bin/local-script → ./bin/local-script; preserved run lint and run script.
• node scripts/run-oxlint.mjs <changed files> — clean.

The P1 real-runtime proof gap (cron/chat/CLI) is still outstanding; I do not have a way to capture that locally without a live agent run.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

P2 v3 follow-up landed in de8e9e3 — addresses the package-script preservation gap.

Heuristic change: strip run only when the rest carries a binary plus arguments (the summarizeKnownExec catch-all shape). Single-token run <X> summaries survive because package-script names (npm run dev, npm run custom-script, npm run start) are indistinguishable from bare-binary invocations (run python3) at strip time — both are run <single-token>. Keeping run in that ambiguous shape preserves the package-script context in the failure message.

This still collapses the originally reported case run python3 /path/to/script.py → python3 /path/to/script.py (catch-all with args), and now also collapses run my-custom-tool --flag value → my-custom-tool --flag value. Single-token labels preserved: run dev, run tests, run build, run lint, run script, run custom-script.

Proof:

• node_modules/.bin/vitest run src/agents/tool-display.test.ts src/agents/embedded-agent-runner/run/payloads.errors.test.ts src/agents/embedded-agent-runner/run/payloads.test.ts — 137/137 pass.
• New coverage in tool-display.test.ts: run dev / run custom-script preserved; multi-token generic binaries stripped.

The P1 real-runtime proof gap (cron/chat/CLI) is still outstanding; I cannot capture that locally without a live agent run.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Closing this as superseded by the canonical shared formatting fix in #97511, landed as 3e90125. #97511 covers the broader payload warning boundary with green proof, so keeping this narrower overlapping PR open would just split the fix path. Thanks for the work here.