The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to aggregate results from various firmware analysis tools (such as binwalk, Ghidra, and QEMU) into a structured report. This ingestion of external, untrusted data creates a potential for indirect prompt injection if malicious payloads are embedded within the analyzed firmware or tool outputs.
Ingestion points: Processes output from firmware extraction, static analysis, reverse engineering, and emulation tools.
Boundary markers: The skill does not define strict delimiters to distinguish between reporting instructions and the data being analyzed.
Capability inventory: The agent can write files, execute system commands (e.g., pandoc, binwalk), and perform network requests through generated proof-of-concept scripts.
Sanitization: There are no instructions for sanitizing or escaping the data ingested from analysis tools before inclusion in the final report.
[DYNAMIC_EXECUTION]: The skill provides templates for generating Python proof-of-concept scripts to validate findings.
Evidence: The finding_template.md and SKILL.md contain Python scripts using the requests library for automated vulnerability verification.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various third-party security tools and system utilities to perform analysis and generate reports.
Evidence: Instructions include running binwalk, readelf, ghidra (headless mode), qemu-arm, tcpdump, and pandoc for PDF generation.

firmware-security-reports