The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the use of the Frida CLI to perform runtime analysis of binaries. These operations are limited to a set of included, reviewable JavaScript scripts and are executed against targets specified by the user within a headless environment.
[SAFE]: The skill incorporates robust data handling policies to prevent accidental information disclosure. It mandates that all raw runtime artifacts remain in a local .work directory and requires manual redaction or generalization of data before it is included in version-controlled Markdown summaries.
[PROMPT_INJECTION]: Analysis of the indirect prompt injection surface identifies that while the skill ingests data from external processes (target binaries), it implements multiple layers of mitigation.
Ingestion points: Runtime state and function parameters captured via Frida hooks in scripts like signature-analysis.js and call-tree-trace.js.
Boundary markers: Usage of a standardized capture manifest (templates/frida-capture-manifest.md) and explicit audit gates to verify captured data before it is moved to the evidence phase.
Capability inventory: Frida instrumentation (Interceptor.attach) is used to monitor function calls within the scope of the analyzed binary.
Sanitization: Non-negotiable constraints in SKILL.md and planning-brief.md require the redaction or generalization of raw values in all tracked outputs.
[SAFE]: No indicators of persistence, privilege escalation, or dynamic code loading from untrusted sources were found. The skill maintains a clear and consistent scope for security research and reverse engineering tasks.

headless-ghidra-frida-runtime-injection