fix(images): skip CLI image cache refs by TurboTheTurtle · Pull Request #87501 · openclaw/openclaw

TurboTheTurtle · 2026-05-28T04:07:53Z

Fixes the sticky-image reattachment portion of #86687.

Summary

skip OpenClaw-written .openclaw-cli-images and openclaw-cli-images paths during prompt image-reference detection
cover direct parser behavior for prior transcript cache paths, temporary cache paths, and file URLs
add a CLI helper regression proving prior .openclaw-cli-images prompt text is not reloaded through loadImageFromRef

Scope note

This intentionally does not change the group-chat selective-reply policy. The issue review marks that half as needing a maintainer/product decision, while the sticky-image loop has a narrow source-proven fix.

Real behavior proof

Behavior or issue addressed: Prior OpenClaw-written CLI image cache paths in prompt history should not be detected or loaded as fresh user image references on later turns.

Real environment tested: Local OpenClaw checkout at /Users/andy/openclaw/openclaw-86687, branch fix/skip-cli-image-cache-86687, dependencies installed with pnpm install --frozen-lockfile.

Exact steps or command run after this patch: Ran pnpm exec tsx -e ... against the real detectImageReferences and loadPromptRefImages helpers using prior prompt text containing /workspace/.openclaw-cli-images/stale.png, then ran ClawSweeper's requested surrounding test suite.

Evidence after fix:

{
  "refs": [
    {
      "raw": "/workspace/current.png",
      "type": "path",
      "resolved": "/workspace/current.png"
    }
  ],
  "cacheOnlyLoadCount": 0,
  "cachePathDetected": false
}

Additional command output:

node scripts/run-vitest.mjs src/agents/cli-runner.helpers.test.ts src/agents/embedded-agent-runner/run/images.test.ts src/auto-reply/reply/groups.test.ts src/config/silent-reply.test.ts --reporter dot -> 6 test files passed, 132 tests passed.

Observed result after fix: The stale .openclaw-cli-images transcript path was ignored, loadPromptRefImages returned zero images for a cache-only prompt, and an unrelated normal image path remained detectable.

What was not tested: I did not run a live Discord channel replay; the fix is scoped to the shared prompt image-reference detector used by the CLI image reload path.

Author attribution

If this PR is squash-merged or reworked, please preserve author attribution for Andy Ye <35905412+TurboTheTurtle@users.noreply.github.com> or include:

Co-authored-by: Andy Ye <35905412+TurboTheTurtle@users.noreply.github.com>

clawsweeper · 2026-05-28T04:10:15Z

Codex review: passed. Reviewed May 28, 2026, 12:35 AM ET / 04:35 UTC.

Summary
The PR updates image-reference detection to ignore OpenClaw CLI image cache directories and adds parser/helper regressions for cache-only prompt text.

PR surface: Source +13, Tests +45. Total +58 across 3 files.

Reproducibility: yes. Current main source shows cache-path prompt text flows from prepareCliPromptImagePayload through loadPromptRefImages into detectImageReferences, and the detector currently has no CLI cache-path exclusion.

Review metrics: none identified.

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🦞 diamond lobster
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

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

Next step before merge

[P2] No repair lane is needed; the submitted PR is focused, automerge-opted, and has no actionable review findings.

Security
Cleared: The diff only changes local image-reference filtering and focused tests; it adds no dependency, workflow, credential, permission, or supply-chain surface.

Review details

Best possible solution:

Land the scoped detector/test fix after required merge gates, while keeping the separate group-chat selective-reply policy work tracked on #86687.

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

Yes. Current main source shows cache-path prompt text flows from prepareCliPromptImagePayload through loadPromptRefImages into detectImageReferences, and the detector currently has no CLI cache-path exclusion.

Is this the best way to solve the issue?

Yes for the sticky-image half. Skipping the internal CLI cache directories in the shared detector is a narrow fix with direct parser/helper coverage; the group-chat selective-reply half remains a separate maintainer/product decision.

AGENTS.md: found and applied where relevant.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 771ddcf1846f.

Label changes

Label changes:

add rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
add status: 🚀 automerge armed: This PR is in ClawSweeper's automerge lane. Sufficient (live_output): The PR body supplies after-fix live helper output showing stale CLI cache refs ignored while a normal image path remains detectable, plus focused test-suite output.
remove rating: 🐚 platinum hermit: Current PR rating is rating: 🦞 diamond lobster, so this older rating label is no longer current.
remove status: 👀 ready for maintainer look: Current PR status label is status: 🚀 automerge armed.

Label justifications:

P1: The PR fixes the P1 sticky-image reattachment regression that can repeatedly leak stale image context into always-on agent turns.
rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
status: 🚀 automerge armed: This PR is in ClawSweeper's automerge lane. Sufficient (live_output): The PR body supplies after-fix live helper output showing stale CLI cache refs ignored while a normal image path remains detectable, plus focused test-suite output.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body supplies after-fix live helper output showing stale CLI cache refs ignored while a normal image path remains detectable, plus focused test-suite output.

Evidence reviewed

PR surface:

Source +13, Tests +45. Total +58 across 3 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	14	1	+13
Tests	2	45	0	+45
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	3	59	1	+58

What I checked:

Current main fallback path: On current main, prepareCliPromptImagePayload falls back to loadPromptRefImages when no fresh images are supplied, and loadPromptRefImages passes the whole prompt to detectImageReferences, which supports the linked sticky-image loop. (src/agents/cli-runner/helpers.ts:337, 771ddcf1846f)
Current main detector behavior: The current detector accepts path and file URL image refs without any .openclaw-cli-images or openclaw-cli-images exclusion, so the reported cache-path prompt text is source-reproducible on main. (src/agents/embedded-agent-runner/run/images.ts:302, 771ddcf1846f)
PR implementation: The branch adds isOpenClawCliImageCachePath and applies it before adding normal path refs and file URL refs, matching the requested sticky-image hot patch without changing config or provider routing. (src/agents/embedded-agent-runner/run/images.ts:102, 6c7ce38a539c)
PR regression coverage: The patch adds detector tests for transcript, temporary cache, and file URL cache paths, plus a CLI helper test proving cache-only prompt text does not call loadImageFromRef or image sanitization. (src/agents/cli-runner.helpers.test.ts:39, 6c7ce38a539c)
Contributor real behavior proof: The PR body includes after-fix helper output where only /workspace/current.png is detected, cacheOnlyLoadCount is 0, and cachePathDetected is false; it also reports the focused surrounding test command passing 132 tests. (6c7ce38a539c)
CI and proof checks: GitHub check-run data for the PR head shows relevant node lanes succeeding and the real behavior proof check succeeding after the proof update; no blocking review comments were present. (6c7ce38a539c)

Likely related people:

steipete: GitHub path history shows bb46b79d3c1479f194a90afcf3dd69a1858a7898 internalized the current image detector and related tests, and recent path history shows repeated CLI/agent image work by this author. (role: feature-history owner; confidence: high; commits: bb46b79d3c14, b9f975b64e80, 42033929d4d0; files: src/agents/embedded-agent-runner/run/images.ts, src/agents/embedded-agent-runner/run/images.test.ts, src/agents/cli-runner/helpers.ts)
mbelinky: Recent merged work in 7299c5695317f74ddbe4a1a68efafd723a38c6c1 changed src/agents/cli-runner/helpers.ts around workspace/cwd handling, adjacent to the cache-path and workspace-scope behavior this PR touches. (role: recent adjacent contributor; confidence: medium; commits: 7299c5695317; files: src/agents/cli-runner/helpers.ts)
Shakker: Earlier commits bbd3ccf8e68a48c3ffa77e1f7de56c7c31a77f80 and 5995c1b4a3a439677710f107b407a7a44f8ed909 added CLI image path reuse and argv coverage in the helper test surface extended by this PR. (role: adjacent test contributor; confidence: medium; commits: bbd3ccf8e68a, 5995c1b4a3a4; files: src/agents/cli-runner.helpers.test.ts)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

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.

clawsweeper · 2026-05-28T04:22:12Z

ClawSweeper PR egg: ✨ hatched 💎 rare Cosmic Proofling. Rarity: 💎 rare. Trait: guards the happy path.

Details

Share on X: post this hatch
Copy: My PR egg hatched a 💎 rare Cosmic Proofling in ClawSweeper.
Hatchability:

Merged PRs are hatchable.
Open PRs are hatchable when they are status: 👀 ready for maintainer look, status: 🚀 automerge armed, or labeled clawsweeper:automerge.
Closed unmerged PRs are hatchable only when one of those hatchable labels is still present in the durable record.

About:

Eggs appear after real-behavior proof passes. They are collectible flavor only.
Review momentum changes the shell state: follow-up work warms it, re-review makes it wobble, and a clean final review lets it hatch.
The hatch is seeded from this repository and PR number, so the same PR keeps the same creature; the reviewed head SHA can only change safe visual details.
Rarity is just collectible sparkle: 🥚 common, 🌱 uncommon, 💎 rare, ✨ glimmer, and 🌈 legendary.

Takhoffman · 2026-05-28T04:26:04Z

@clawsweeper visualize

clawsweeper · 2026-05-28T04:26:07Z

🦞👀
ClawSweeper visual brief is being prepared.

I queued a read-only visual pass. It will create or update one marker-backed visual brief comment and will not trigger close, merge, repair, label, or branch changes.

Lens: auto

clawsweeper · 2026-05-28T04:27:42Z

Source: #87501 (comment)
Visual model: gpt-5.5, reasoning low.

Visual brief

Requested lens: state
Advisory only: maintainers remain the final judges.

PR: #87501
Related issue: #86687

Before: stale image cache path becomes live image input again

Turn N
  user attaches image
        |
        v
  OpenClaw writes cache path
  /workspace/.openclaw-cli-images/stale.png
        |
        v
  transcript stores path text
        |
        v
Turn N+1, no new image
  prompt history includes prior cache path
        |
        v
  detectImageReferences()
        |
        v
  🐛 matches OpenClaw-written cache path
        |
        v
  loadPromptRefImages()
        |
        v
  ❌ stale image bytes reattached

After: OpenClaw-written cache refs are filtered

Turn N+1 prompt text
+----------------------------------------------+
| prior transcript cache path                  |
| /workspace/.openclaw-cli-images/stale.png    |
|                                              |
| current user image path                      |
| /workspace/current.png                       |
+----------------------------------------------+
           |
           v
 detectImageReferences()
           |
           +--> 💾 .openclaw-cli-images/...  -> ✅ ignored
           |
           +--> user/current image path       -> ✅ preserved

Patch shape

src/agents/embedded-agent-runner/run/images.ts
  + isOpenClawCliImageCachePath()
  + skip .openclaw-cli-images and openclaw-cli-images path parts
  + apply skip to normal paths and file:// paths

Tests added
  ✅ direct transcript cache path ignored
  ✅ temporary openclaw-cli-images path ignored
  ✅ file:// cache URL ignored
  ✅ CLI helper does not call loadImageFromRef for cache-only prompt

Scope boundary

Related issue #86687 has two reported regressions:

+-------------------------------+-------------------------------+
| Sticky image reattachment     | ✅ addressed by this PR        |
| Group-chat selective replies  | ⚠️ intentionally unchanged     |
+-------------------------------+-------------------------------+

Proof supplied

Input proof:
  prompt contains /workspace/.openclaw-cli-images/stale.png
  prompt also contains /workspace/current.png

Observed after fix:
  refs = [ /workspace/current.png ]
  cacheOnlyLoadCount = 0
  cachePathDetected = false

Reported test run:
  node scripts/run-vitest.mjs ... --reporter dot
  ✅ 6 files passed
  ✅ 132 tests passed

Legend: ✅ expected/proven; ❌ broken/dropped path; ⚠️ unresolved tradeoff; 🐛 confirmed pre-fix bug; 💾 persisted disk/cache state.

Maintainer ruling

Benefit: Stops OpenClaw-written CLI image cache paths from being re-detected as fresh user image references on later turns, while preserving ordinary image-path detection.

Risk: Path-part filtering may also ignore a user-authored image path if it intentionally lives inside a directory named .openclaw-cli-images or openclaw-cli-images.

Proof needed: Maintainer/CI confirmation on the submitted branch, plus any desired review that the directory-name filter is the right product boundary.

Recommended next action: Maintainer review of the narrow image-cache filter and its tests.

Question presented: Is skipping any path segment named .openclaw-cli-images or openclaw-cli-images the intended boundary for OpenClaw-owned image cache refs?

Takhoffman · 2026-05-28T04:29:02Z

@clawsweeper auto merge

clawsweeper · 2026-05-28T04:29:04Z

🦞🔧
ClawSweeper picked up the repair feedback.

Source: clawsweeper[bot]
Feedback: trusted ClawSweeper review contains P-severity findings (sha=6c7ce38a539c158ef4a96c82129711eab8536ece)
Action: repair worker queued. Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586
Model: gpt-5.5

I will update this PR branch, or open a safe credited replacement, if the repair worker finds a narrow fix.

Automerge progress:

2026-05-28 04:29:37 UTC review queued 6c7ce38a539c (queued)

2026-05-28 04:29:50 UTC review queued 6c7ce38a539c (queued)

2026-05-28 04:35:31 UTC review requested repair 6c7ce38a539c (trusted ClawSweeper review contains P-severity findings (sha=6c7ce38a539c158ef4...)

2026-05-28 04:35:56 UTC repair queued 6c7ce38a539c (autonomous) Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586

2026-05-28 04:37:09 UTC repair started (running) in 0s Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586 automerge-openclaw-openclaw-87501

2026-05-28 04:37:26 UTC validation plan (passed) in 17s Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586 pnpm check:changed; pnpm lint; pnpm check:test-types

2026-05-28 04:37:41 UTC Codex write preflight (passed) in 33s Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586 danger-full-access

2026-05-28 04:46:03 UTC Codex edit 1 5bfba6ae4a3a (complete) in 8m 55s Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586 exit 0

2026-05-28 04:54:12 UTC validation and review 1 c166f8b74d9a (base moved) in 17m 3s Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586 rebased

2026-05-28 04:54:48 UTC repair finished c166f8b74d9a (opened) in 17m 40s Run: https://github.com/openclaw/clawsweeper/actions/runs/26554925586 open_fix_pr

Takhoffman · 2026-05-28T04:29:10Z

@clawsweeper automerge

clawsweeper · 2026-05-28T04:29:12Z

🦞👀
ClawSweeper picked this up.

Command router queued. I will update this comment with the next step.

clawsweeper · 2026-05-28T04:54:46Z

ClawSweeper 🐠 reef update

Thanks for the contribution. The source branch was not safely writable by ClawSweeper, so it opened a replacement PR and kept the credit trail visible.

Why replacement: ClawSweeper could not update the source PR branch directly; GitHub did not grant sufficient push rights to the bot for that branch.
Replacement PR: #87523
Why close: this run explicitly closes the superseded source PR after the credited replacement PR is open, so review continues in one place.
Closing this source PR only because source-PR closing was explicitly enabled for this run.
The replacement PR carries the original credit trail forward.
Co-author credit kept:

@TurboTheTurtle: Co-authored-by: Andy Ye 35905412+TurboTheTurtle@users.noreply.github.com

fish notes: model gpt-5.5, reasoning high; reviewed against c166f8b.

fix(images): skip CLI image cache refs

6c7ce38

openclaw-barnacle Bot added agents Agent runtime and tooling size: S triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 28, 2026

openclaw-barnacle Bot added proof: supplied External PR includes structured after-fix real behavior proof. and removed triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 28, 2026

clawsweeper Bot mentioned this pull request May 28, 2026

fix(images): skip CLI image cache refs #87523

Merged

clawsweeper Bot closed this May 28, 2026

TurboTheTurtle deleted the fix/skip-cli-image-cache-86687 branch May 28, 2026 05:47

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(images): skip CLI image cache refs#87501

fix(images): skip CLI image cache refs#87501
TurboTheTurtle wants to merge 1 commit into
openclaw:mainfrom
TurboTheTurtle:fix/skip-cli-image-cache-86687

TurboTheTurtle commented May 28, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 28, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

Takhoffman commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

Takhoffman commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026 •

edited

Loading

Uh oh!

Takhoffman commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

TurboTheTurtle commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Scope note

Real behavior proof

Author attribution

Uh oh!

clawsweeper Bot commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

Takhoffman commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

clawsweeper Bot commented May 28, 2026

Visual brief

Maintainer ruling

Uh oh!

Takhoffman commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Takhoffman commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

clawsweeper Bot commented May 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

TurboTheTurtle commented May 28, 2026 •

edited

Loading

clawsweeper Bot commented May 28, 2026 •

edited

Loading

clawsweeper Bot commented May 28, 2026 •

edited

Loading

clawsweeper Bot commented May 28, 2026 •

edited

Loading