The Agent Skills Directory

[SAFE]: The skill implements strong security guardrails by requiring explicit authorization checks before performing reverse engineering tasks. It mandates reading a precedent journal to ensure the operation is authorized.
[SAFE]: Binary modification is gated behind a 'write mode' confirmation. The instructions explicitly warn the agent to use read-only mode by default and only switch to write mode (r2 -w) after notifying the user of potential risks and suggesting backups.
[SAFE]: The reconnaissance scripts (recon.ps1 and recon.sh) use standard, well-documented radare2 utilities (rabin2, r2) to extract metadata. External dependencies are handled through an on-demand bootstrap mechanism targeting official GitHub releases, which is a standard deployment pattern for this toolset.
[SAFE]: The skill correctly scopes its environment by limiting tool access through allowed-tools logic and provides specific paths for tool discovery, preventing command injection or path hijacking.
[SAFE]: Indirect prompt injection risks are mitigated by the 'imports gatekeeper' rule, which forces the agent to document all imported APIs (Evidence-IAT) before proceeding with analysis, ensuring that malicious code behavior is identified early in the workflow.

radare2