CVE-2026-35632

OpenClaw before 2026.3.24 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.

OpenClaw before 2026.3.24 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.

Summary

The patch for CVE-2026-32013 introduced symlink resolution and workspace boundary enforcement for agents.files.get and agents.files.set. However, two other handlers in the same file (agents.create and agents.update) still use raw fs.appendFile on the IDENTITY.md file without any symlink containment check. An attacker who can place a symlink in the agent workspace can hijack the IDENTITY.md path to append attacker-controlled content to arbitrary files on the system.

Details

In src/gateway/server-methods/agents.ts, the agents.create handler constructs the identity path and appends agent metadata without verifying symlinks:

// agents.create — line 283-291
const identityPath = path.join(workspaceDir, DEFAULT_IDENTITY_FILENAME);
const lines = [
  "",
  `- Name: ${safeName}`,
  ...(emoji ? [`- Emoji: ${sanitizeIdentityLine(emoji)}`] : []),
  ...(avatar ? [`- Avatar: ${sanitizeIdentityLine(avatar)}`] : []),
  "",
];
await fs.appendFile(identityPath, lines.join("\n"), "utf-8"); // ← NO SYMLINK CHECK

The agents.update handler has the same issue at line 348-349:

// agents.update — line 348-349
const identityPath = path.join(workspace, DEFAULT_IDENTITY_FILENAME);
await fs.appendFile(identityPath, `\n- Avatar: ${sanitizeIdentityLine(avatar)}\n`, "utf-8"); // ← NO SYMLINK CHECK

fs.appendFile follows symlinks by default. If the IDENTITY.md file in the workspace is a symlink pointing to a sensitive file (e.g., /etc/crontab, ~/.bashrc, or ~/.ssh/authorized_keys), calling agents.create will append the agent identity metadata to that file.

The ensureAgentWorkspace function (called at line 274 before the append) uses exclusive-create mode (flag: 'wx') for IDENTITY.md. If a symlink already exists at that path, the EEXIST error is silently caught, and the subsequent fs.appendFile follows the symlink.

Attack flow:

1. Attacker plants symlink: workspace/IDENTITY.md → /etc/crontab
2. ensureAgentWorkspace skips creation (EEXIST from symlink)
3. fs.appendFile follows symlink → writes to /etc/crontab
4. Attacker-controlled content (name, emoji, avatar) injected into crontab → RCE

PoC

Prerequisites: Docker and Python 3 installed.

Step 1: Build and start the test environment.

cd llm-enhance/cve-finding/RCE/CVE-2026-32013-identity-appendFile-variant-exp/
docker compose up -d --build
sleep 3

Step 2: Run the exploit.

python3 poc_exploit.py

This script:

1. Plants a symlink IDENTITY.md → /etc/target-file.txt inside the agent workspace
2. Calls the agents.create API endpoint via HTTP POST
3. Verifies that the agent identity metadata was appended to /etc/target-file.txt

Step 3: Run the control experiment.

python3 control-patched_realpath.py

Step 4: Cleanup.

docker compose down

Log of Evidence

Exploit output:

=== CVE-2026-32013 Variant: Symlink Traversal via IDENTITY.md appendFile ===
[*] Planting symlink: IDENTITY.md -> /etc/target-file.txt
[*] Symlink: lrwxrwxrwx 1 root root 20 /workspaces/evil-agent/IDENTITY.md -> /etc/target-file.txt
[*] Original /etc/target-file.txt: ORIGINAL_SENSITIVE_CONTENT

[*] Calling agents.create with name='evil-agent'...
[*] API response: {'ok': True, 'agentId': 'evil-agent', 'workspace': '/workspaces/evil-agent'}

[*] /etc/target-file.txt after exploit:
    ORIGINAL_SENSITIVE_CONTENT

- Name: evil-agent
- Emoji: 💀
- Avatar: evil.png

[+] SUCCESS! Symlink traversal confirmed.
[+] fs.appendFile followed IDENTITY.md symlink and wrote to /etc/target-file.txt
[+] Attacker-controlled content injected into arbitrary file.

Control output:

=== CONTROL: Patched agents.create blocks symlink traversal ===
[*] Planting symlink: IDENTITY.md -> /etc/target-file.txt
[*] Original /etc/target-file.txt: ORIGINAL_SENSITIVE_CONTENT

[*] Calling PATCHED agents.create with name='safe-agent'...
[*] API…

Duplicate Advisory

This advisory has been withdrawn because it is a duplicate of GHSA-7xr2-q9vf-x4r5. This link is maintained to preserve external references.

Original Description

OpenClaw through 2026.2.22 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.

OpenClaw before 2026.3.24 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.

OpenClaw before 2026.3.24 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.

Summary

The patch for CVE-2026-32013 introduced symlink resolution and workspace boundary enforcement for agents.files.get and agents.files.set. However, two other handlers in the same file (agents.create and agents.update) still use raw fs.appendFile on the IDENTITY.md file without any symlink containment check. An attacker who can place a symlink in the agent workspace can hijack the IDENTITY.md path to append attacker-controlled content to arbitrary files on the system.

Details

In src/gateway/server-methods/agents.ts, the agents.create handler constructs the identity path and appends agent metadata without verifying symlinks:

// agents.create — line 283-291
const identityPath = path.join(workspaceDir, DEFAULT_IDENTITY_FILENAME);
const lines = [
  "",
  `- Name: ${safeName}`,
  ...(emoji ? [`- Emoji: ${sanitizeIdentityLine(emoji)}`] : []),
  ...(avatar ? [`- Avatar: ${sanitizeIdentityLine(avatar)}`] : []),
  "",
];
await fs.appendFile(identityPath, lines.join("\n"), "utf-8"); // ← NO SYMLINK CHECK

The agents.update handler has the same issue at line 348-349:

// agents.update — line 348-349
const identityPath = path.join(workspace, DEFAULT_IDENTITY_FILENAME);
await fs.appendFile(identityPath, `\n- Avatar: ${sanitizeIdentityLine(avatar)}\n`, "utf-8"); // ← NO SYMLINK CHECK

fs.appendFile follows symlinks by default. If the IDENTITY.md file in the workspace is a symlink pointing to a sensitive file (e.g., /etc/crontab, ~/.bashrc, or ~/.ssh/authorized_keys), calling agents.create will append the agent identity metadata to that file.

The ensureAgentWorkspace function (called at line 274 before the append) uses exclusive-create mode (flag: 'wx') for IDENTITY.md. If a symlink already exists at that path, the EEXIST error is silently caught, and the subsequent fs.appendFile follows the symlink.

Attack flow:

1. Attacker plants symlink: workspace/IDENTITY.md → /etc/crontab
2. ensureAgentWorkspace skips creation (EEXIST from symlink)
3. fs.appendFile follows symlink → writes to /etc/crontab
4. Attacker-controlled content (name, emoji, avatar) injected into crontab → RCE

PoC

Prerequisites: Docker and Python 3 installed.

Step 1: Build and start the test environment.

cd llm-enhance/cve-finding/RCE/CVE-2026-32013-identity-appendFile-variant-exp/
docker compose up -d --build
sleep 3

Step 2: Run the exploit.

python3 poc_exploit.py

This script:

1. Plants a symlink IDENTITY.md → /etc/target-file.txt inside the agent workspace
2. Calls the agents.create API endpoint via HTTP POST
3. Verifies that the agent identity metadata was appended to /etc/target-file.txt

Step 3: Run the control experiment.

python3 control-patched_realpath.py

Step 4: Cleanup.

docker compose down

Log of Evidence

Exploit output:

=== CVE-2026-32013 Variant: Symlink Traversal via IDENTITY.md appendFile ===
[*] Planting symlink: IDENTITY.md -> /etc/target-file.txt
[*] Symlink: lrwxrwxrwx 1 root root 20 /workspaces/evil-agent/IDENTITY.md -> /etc/target-file.txt
[*] Original /etc/target-file.txt: ORIGINAL_SENSITIVE_CONTENT

[*] Calling agents.create with name='evil-agent'...
[*] API response: {'ok': True, 'agentId': 'evil-agent', 'workspace': '/workspaces/evil-agent'}

[*] /etc/target-file.txt after exploit:
    ORIGINAL_SENSITIVE_CONTENT

- Name: evil-agent
- Emoji: 💀
- Avatar: evil.png

[+] SUCCESS! Symlink traversal confirmed.
[+] fs.appendFile followed IDENTITY.md symlink and wrote to /etc/target-file.txt
[+] Attacker-controlled content injected into arbitrary file.

Control output:

=== CONTROL: Patched agents.create blocks symlink traversal ===
[*] Planting symlink: IDENTITY.md -> /etc/target-file.txt
[*] Original /etc/target-file.txt: ORIGINAL_SENSITIVE_CONTENT

[*] Calling PATCHED agents.create with name='safe-agent'...
[*] API…

Duplicate Advisory

This advisory has been withdrawn because it is a duplicate of GHSA-7xr2-q9vf-x4r5. This link is maintained to preserve external references.

Original Description

OpenClaw through 2026.2.22 contains a symlink traversal vulnerability in agents.create and agents.update handlers that use fs.appendFile on IDENTITY.md without symlink containment checks. Attackers with workspace access can plant symlinks to append attacker-controlled content to arbitrary files, enabling remote code execution via crontab injection or unauthorized access via SSH key manipulation.