firmware-static-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process and analyze content from untrusted ELF binaries using tools like
strings,readelf, andobjdump(SKILL.md, Steps 2, 3, 4, 7). This content is ingested into the agent context without sanitization or explicit boundary markers. - Ingestion points: The agent reads the output of
strings,readelf, andobjdumpcommands performed on untrusted binary files (SKILL.md). - Boundary markers: Absent; there are no instructions to use specific delimiters or to treat the ingested tool output as potentially untrusted instructions.
- Capability inventory: The skill allows the agent to execute multiple shell commands including
file,strings,readelf,objdump,xxd,strip,nm,ldd, andbinwalk(SKILL.md, toolchain.md). - Sanitization: Absent; the instructions do not include any steps to sanitize or filter the output from binary analysis tools before the agent processes it.
- [COMMAND_EXECUTION]: The skill facilitates the systematic execution of multiple command-line utilities for binary reconnaissance. While these are standard tools for static analysis, the reliance on shell execution across all workflow steps increases the overall attack surface. Notably, the toolchain reference includes
ldd, which executes the binary's dynamic linker, posing a risk if used on untrusted files; however, the skill provides an explicit warning against this practice. (Found in SKILL.md and references/toolchain.md).
Audit Metadata