T06 · System Persistence
- Location
SKILL.md:133- Finding
Hidden Cross-Session Execution Through the Windows Startup Folder
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent local Ollama/OpenClaw memory setup guide, but it asks users to run unverified downloaded binaries and create hidden Windows startup persistence without enough safeguards.
Review before installing. Use only a verified Ollama release, check hashes and signatures, avoid the hidden Startup-folder autostart unless you explicitly need it, and treat OpenClaw memory backups as sensitive private data. Do not run the referenced rebuild script unless you have verified its source and contents.
SKILL.md:133Hidden Cross-Session Execution Through the Windows Startup Folder
SKILL.md:44Mutable External Binary Retrieval Followed by Local Execution
SKILL.md:234Unpinned Python Dependencies and Invocation of an Unshipped Script
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
python -c "import requests; r = requests.post('http://localhost:11434/api/embeddings', json={'model':'nomic-embed-text','prompt':'test'}); print(f'Dim: {len(r.json()["embedding"])}')"
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
python -c "import requests; r = requests.post('http://localhost:11434/api/embeddings', json={'model':'nomic-embed-text','prompt':'test'}); print(f'Dim: {len(r.json()["embedding"])}')"
Code issues a request to a loopback, link-local, or private-range host. This can reach internal services not meant to be exposed and is a common SSRF pivot.
python -c "import requests; r = requests.post('http://localhost:11434/api/embeddings', json={'model':'nomic-embed-text','prompt':'test'}); print(f'Dim: {len(r.json()["embedding"])}')"
The guide explicitly instructs users to back up and restore memory directories that likely contain conversation logs, extracted memories, and user profile data, but it does not warn that these artifacts may be highly sensitive. In an enterprise setting, this omission can lead to insecure copying, broad filesystem exposure, or migration of personal data without proper access controls or retention review.
All natural-language instructions in the file are presented in Chinese, with no opt-in, alternative locale, or justification that the skill is region- or language-specific. Under the policy rule for natural-language locale constraints, forcing a single language without user choice can be a policy violation.
The package description is written entirely in Chinese, which imposes a language choice in a user-facing skill descriptor without indicating any language option or opt-in. This can violate a language/locale policy when users are not given a choice or told the skill is intentionally region- or language-specific.
No suspicious patterns detected.