host-observer
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external observer jobs, which provides a surface for untrusted data to enter the agent context and influence behavior.
- Ingestion points: Observer jobs from Grok Bot and claude-mem as described in SKILL.md.
- Boundary markers: The skill does not specify any delimiters or instructions to isolate the untrusted data from its primary instructions.
- Capability inventory: The skill specifies writing to the local file system ('outbox/{id}.txt') and performing network POST requests ('/api/memory/save').
- Sanitization: No sanitization or validation procedures are mentioned for processing the input data.
- [DATA_EXFILTRATION]: The skill includes instructions to perform network POST requests and write to the local file system. These capabilities could be leveraged to exfiltrate or expose data if the processed jobs contain malicious instructions.
- [COMMAND_EXECUTION]: The skill instructs the agent to write to 'outbox/{id}.txt', which is a file system operation that could be abused for path traversal or unauthorized file creation if the ID or content is not validated.
Audit Metadata