debugger
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted binary data from the filesystem using functions like disasm_at and the bytes table. Maliciously crafted binaries could contain instructions or strings designed to manipulate the agent's logic. Ingestion points: bytes table and disassembly functions in SKILL.md. Capability inventory: The skill has access to Bash (via allowed-tools) and can modify the binary via patch_byte. Boundary markers: None identified. Sanitization: None identified.
- [DATA_EXFILTRATION]: The load_file_bytes function documented in the skill allows reading any file on the host system into the agent's context. While this is a common debugger feature, it presents a risk of sensitive data exposure if used to target system configuration or credential files.
Audit Metadata