The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of disassembly and pseudocode extracted from binary files to facilitate comparison by an LLM.
Ingestion points: The variables {disasm_for_reference}, {procedure_for_reference}, {disasm_code}, and {procedure} in references/prompt-template.md serve as entry points for data that could contain embedded instructions.
Boundary markers: The prompt template uses Markdown triple-backtick blocks (c ... ) to delimit the code sections, which provides basic structural separation but does not prevent sophisticated injection.
Capability inventory: The provided Python script skeleton uses httpx to send data to external LLM providers and pyyaml to parse the results, which are then used to generate renaming commands for IDA Pro.
Sanitization: The script correctly uses yaml.safe_load() to parse the LLM's output, mitigating risks associated with untrusted YAML data execution.
[EXTERNAL_DOWNLOADS]: The skill documentation mentions dependencies on external Python packages and interaction with remote AI services.
Evidence: Instructions in SKILL.md specify installing pyyaml via pip. The script in references/prompt-template.md uses the httpx library and targets official API endpoints for DeepSeek, OpenAI, and Anthropic.
Note: These targets are well-known technology services and the dependencies are standard for the skill's stated purpose.

binary-diff