GHSA-j425-whc4-4jgc: OpenClaw system.run Env Override Filtering Allowed Dangerous Helper-Command Pivots
Keywords: GHSA-j425-whc4-4jgc, OpenClaw vulnerability, system.run env override, sanitizeHostExecEnv bypass, GIT_SSH_COMMAND, helper-command pivot, AI agent security, env override filtering, allowlist bypass, OpenClaw security advisory
Table of Contents
Overview
OpenClaw system.run env override filtering allowed dangerous helper-command pivots — The env sanitization in src/infra/host-env-security.ts blocked only a narrow set of override-only environment variables, allowing a tool-enabled caller to steer allowlisted tools through helper-command or config-loading variables like GIT_SSH_COMMAND, editor/pager hooks, and GIT_CONFIG_* / NPM_CONFIG_*.
OpenClaw (187K+ GitHub stars) is a widely deployed personal AI assistant. Its system.run capability allows the agent to execute shell commands on the host machine. Approval evaluation was tied to the reviewed binary/argv, but the launched process could still inherit attacker-controlled env overrides that changed helper-command execution or config resolution — meaning the actual spawned-process behavior diverged from what was approved.
Discovered by: Kai Aizen (@SnailSploit) & @tdjackey
Published: March 8, 2026
CVSS Score: 6.3 (Moderate)
CWE: CWE-15 (External Control of System or Configuration Setting) + CWE-639 (Authorization Bypass Through User-Controlled Key)
Package: openclaw (npm)
Patched in: 2026.3.7
Vulnerability Details
Description
Before the fix, src/infra/host-env-security.ts blocked only a narrow set of override-only environment variables. Dangerous request-scoped overrides such as GIT_SSH_COMMAND and prefix families such as GIT_CONFIG_* and NPM_CONFIG_* could still survive sanitizeSystemRunEnvOverrides(...) / sanitizeHostExecEnv(...) and reach the spawned process.
This mattered for system.run allowlist and approval flows because approval evaluation was tied to the reviewed binary/argv, while the launched process could still inherit attacker-controlled env overrides that changed helper-command execution or config resolution. For allowlisted tools such as git, this allowed behavior outside the reviewed command semantics.
Affected Versions
- Vulnerable: All versions ≤ 2026.3.2
- Patched: Version 2026.3.7
Technical Analysis
The core issue is a semantic mismatch between what was reviewed and what actually executes:
- A caller invokes
system.run with env overrides (e.g., { "GIT_SSH_COMMAND": "<payload>" })
- The allowlist/approval system reviews the binary and argv —
git push origin main looks safe
sanitizeSystemRunEnvOverrides() / sanitizeHostExecEnv() check against the blocklist
GIT_SSH_COMMAND is not blocked — it passes through to the spawned process
- Git invokes the attacker-controlled
GIT_SSH_COMMAND as a helper — executing behavior outside the reviewed command semantics
Dangerous Override Categories
| Category |
Examples |
Effect |
| Helper-command pivots |
GIT_SSH_COMMAND, GIT_ASKPASS, EDITOR, VISUAL |
Direct command execution via tool hooks |
| Config-loading overrides |
GIT_CONFIG_*, NPM_CONFIG_* |
Config resolution hijacking |
| Pager/display hooks |
LESSOPEN, LESSCLOSE, GIT_PAGER |
Command execution on output display |
Vulnerable Code Path
src/infra/host-env-security.ts → sanitizeSystemRunEnvOverrides()
src/infra/host-env-security.ts → sanitizeHostExecEnv()
src/infra/host-env-security-policy → narrow blocklist (insufficient)
Fix
The fix (commit e27bbe4) extends the shared TypeScript and macOS policy to block dangerous override-only exact keys and prefixes while preserving trusted inherited base-environment behavior.
Attack Chain
┌─────────────────────────────────────────────────────────────┐
│ 1. Tool-enabled caller invokes system.run with env │
│ overrides: { GIT_SSH_COMMAND: "<payload>" } │
├─────────────────────────────────────────────────────────────┤
│ 2. Allowlist/approval evaluates the binary + argv │
│ "git push origin main" → approved │
├─────────────────────────────────────────────────────────────┤
│ 3. sanitizeHostExecEnv checks blocklist │
│ GIT_SSH_COMMAND → NOT BLOCKED → passes through │
├─────────────────────────────────────────────────────────────┤
│ 4. Spawned git process inherits the env override │
│ Git invokes GIT_SSH_COMMAND as helper │
├─────────────────────────────────────────────────────────────┤
│ 5. Actual behavior diverges from what was approved │
│ Helper-command executes attacker-controlled payload │
└─────────────────────────────────────────────────────────────┘
Impact
This is a protection-bypass issue. Exploitation requires an already tool-enabled caller who can invoke system.run and supply env overrides. In affected deployments, that caller could:
- Bypass allowlist/approval intent — trigger helper-command execution not represented by the approved command line
- Hijack config resolution — steer tool behavior through config-loading env vars
- Achieve behavior outside reviewed command semantics — the approved
git push becomes something entirely different at runtime
Maintainer severity is set to Moderate because the bug still requires existing execution capability; the vulnerability is the mismatch between reviewed command semantics and actual spawned-process behavior.
Remediation
Immediate Fix
Update OpenClaw to version 2026.3.7 or later:
npm install -g openclaw@latest
Defense in Depth
- Audit
system.run invocations for env override usage
- Prefer allowlist-based env filtering over blocklist approaches
- Monitor subprocess behavior for divergence from approved command semantics
CVSS v3.1 Metrics
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:L
| Metric |
Value |
| Attack Vector |
Network (AV:N) |
| Attack Complexity |
Low (AC:L) |
| Privileges Required |
Low (PR:L) |
| User Interaction |
None (UI:N) |
| Scope |
Unchanged (S:U) |
| Confidentiality |
Low (C:L) |
| Integrity |
Low (I:L) |
| Availability |
Low (A:L) |
Timeline
| Date |
Event |
| 2026-03-03 |
Vulnerability discovered via source code review |
| 2026-03-04 |
Reported via GitHub Security Advisory |
| 2026-03-08 |
Advisory published (GHSA-j425-whc4-4jgc) |
| 2026-03-08 |
Patched in OpenClaw 2026.3.7 |
| 2026-03-09 |
Published to GitHub Advisory Database, reviewed |
References
Contact
Kai Aizen (SnailSploit)
⚠️ Disclaimer: This repository is for educational and authorized security research purposes only. Use responsibly.
📚 Documentation & Author
This project's full writeup, methodology, and related research lives at:
https://snailsploit.com/advisories
Created by Kai Aizen — independent offensive security researcher.
snailsploit.com · Research · Frameworks · GitHub · LinkedIn · ResearchGate · X/Twitter
Same attack. Different substrate.