GHSA-7ff8-xjh3-mgh6: OpenClaw's non-default autoAllowSkills setting could bypass on-miss exec prompt
With the non-default `autoAllowSkills=true`, OpenClaw 2026.2.22-2 and earlier let `./skill-bin` match a skill named `skill-bin` and run without an `ask=on-miss` approval prompt. Default installs are not affected. Upgrade to openclaw 2026.2.23.
Published Mar 3, 2026 · Analysis Oct 31, 2026
What an attacker can do
Skill auto-allow lets commands from installed skills run without a prompt. The allowlist evaluator decided a command satisfied the skills segment by comparing its basename to the skill bin names. A path-scoped command such as ./skill-bin, or an absolute path with the same basename, resolved to skill-bin and matched skillBins.has("skill-bin"), so the prompt was skipped even though the file was not the skill's binary.
The bug needs all three settings: autoAllowSkills=true (default false), system.run with security=allowlist, and ask=on-miss. The attacker, or an agent steered by prompt injection, also needs a way to put an executable with a skill's name where the command can reach it, such as a file in the working directory. The CVSS 4.0 vector is local, low privilege and high attack complexity.
The result is that commands which should have waited for operator confirmation run immediately, with the OpenClaw user's rights and high impact on confidentiality, integrity and availability.
Who is affected
openclaw (npm) versions up to and including 2026.2.22-2, only where autoAllowSkills=true, system.run uses security=allowlist, and ask=on-miss. Default installs are not affected.
npm ls openclawThen check your exec approval config for autoAllowSkills. If it is false or unset, the advisory's conditions are not met.
openclaw
- Ecosystem
- npm
- Affected
- <= 2026.2.22-2
- Fixed in
- 2026.2.23
- Check
npm ls openclaw
How to fix it
Upgrade to openclaw 2026.2.23:
npm install [email protected]Skill auto-allow now requires a pathless invocation token, with no / or \\, and a resolved executable path that matches a trusted entry for that skill bin on the machine where skills run. Normal skill-bin ... calls still work. ./skill-bin and absolute-path basename collisions are denied with an approval-required result. As a stopgap, set autoAllowSkills back to false.
How to tell if you were hit
Review exec logs for commands starting with ./ or an absolute path whose basename equals a skill bin and that ran without a prompt. Check working directories for executables named like your skills. After the upgrade, such calls are denied with an exec.denied event and reason approval-required. Beam records agent shell commands and flags risky ones, which shows what ran without confirmation.
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
Summary
In openclaw versions up to and including 2026.2.22-2, a non-default exec-approval configuration could allow a skill-name collision to bypass an ask=on-miss prompt.
When autoAllowSkills=true, a path-scoped executable such as ./skill-bin could resolve to basename skill-bin, satisfy the skills allowlist segment, and run without prompting for approval.
Affected Packages / Versions
- Package:
npm openclaw - Affected versions:
<= 2026.2.22-2 - Patched versions:
>= 2026.2.23(released)
Configuration Scope (Not Default)
This behavior requires non-default settings and does not affect default installs.
Required conditions:
autoAllowSkills=true(default isfalse)system.runwithsecurity=allowlistask=on-miss
Technical Details
The allowlist evaluator accepted skills satisfaction by bin-name match, so ./skill-bin could match skillBins.has("skill-bin") after resolution.
The fix hardens skill auto-allow matching by requiring:
- a pathless invocation token (no
/or\\), and - a trusted resolved executable path for that skill bin on the machine where skills run.
This preserves normal skill-bin ... behavior while preventing ./<skill-bin> and absolute-path basename collisions from auto-satisfying skills.
Impact
In affected non-default configurations, approval prompts could be skipped for commands that should have required operator confirmation.
Fix Commit(s)
ffd63b7a2c4c6d5aeb4710ef951d5794ad7ad77b(fix(security): trust resolved skill-bin paths in allowlist auto-allow)
OpenClaw thanks @tdjackey for reporting.
Credited to tdjackey.
Questions
- Are default installs affected by this OpenClaw advisory?
No. The advisory says it needs non-default settings:
autoAllowSkills=true,system.runwithsecurity=allowlist, andask=on-miss.- Does normal skill usage still work after the fix?
Yes. Pathless calls like
skill-bin ...that resolve to a trusted path still auto-allow. Only path-scoped invocations and basename collisions are blocked.- Which version fixes it?
openclaw 2026.2.23. Versions up to 2026.2.22-2 are affected. The advisory has no CVE id in the data.
References
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology
More OpenClaw vulnerabilities
All OpenClaw entriesCVE-2026-62197
OpenClaw: Browser CDP discovery could accept blocked WebSocket URLs
medium8.5Oct 21, 2026CVE-2026-35630
OpenClaw: QQBot native approval buttons did not enforce configured approver identity
high8.0Jul 2, 2026CVE-2026-53806
OpenClaw: Combined POSIX shell options could confuse exec revalidation
high8.8Jul 2, 2026CVE-2026-53809
OpenClaw: Embedded runner policy could be confused by provider aliases
medium3.8Jul 2, 2026CVE-2026-53810
OpenClaw's marketplace runtime extension metadata could point at unscanned payloads
high8.8Jul 2, 2026CVE-2026-53811
OpenClaw: Matrix allowFrom could bind to mutable display names
high8.8Jul 2, 2026
