The Agent Skills Directory

[SAFE]: Comprehensive analysis of the instructions and metadata shows no evidence of prompt injection, data exfiltration, obfuscation, or malicious command execution. The skill performs legitimate technical triage for reverse engineering workflows.
[INDIRECT_PROMPT_INJECTION]: The skill involves processing external, potentially untrusted data from binary analysis tools and compiler diffs, which represents an indirect prompt injection surface.
Ingestion points: External compiler diff outputs, jump-table mappings, and symbol data from tools like dumpbin and llvm-nm are processed in SKILL.md.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to treat this external data as untrusted within the provided documentation.
Capability inventory: The skill references capabilities to execute binary analysis tools and interact with complementary reconstruction skills (matcher, permute) in SKILL.md.
Sanitization: The skill does not provide instructions for sanitizing or escaping the data ingested from translation units or binary analysis reports.

wall-identifier