The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from web research, repository code, and local artifacts to generate explanations. Ingestion points: Web search results, external source code files, and user-supplied documents. Boundary markers: The skill uses an 'Evidence Contract' (references/evidence-contract.md) to distinguish between observed facts, corroborated behavior, inference, and unknowns, and includes instructions to 'label material inference and unknowns'. Capability inventory: The skill uses Playwright for rendering checks (scripts/render-check.sh), performs file system reads across repositories, and executes web searches. Sanitization: Instructions in references/visual-explanation.md mandate escaping markup when including code snippets in the generated HTML artifacts.
[EXTERNAL_DOWNLOADS]: The skill utilizes the Playwright CLI, a well-known service, for automated rendering and screenshots of generated explainers. It also references an external GitHub repository (zhaoxuya520/reverse-skill) as a design source in its documentation (references/reverse-core.md).
[DYNAMIC_EXECUTION]: The skill includes a shell script (scripts/render-check.sh) that dynamically executes a Python snippet for URI conversion and invokes the Playwright CLI. Additionally, references/reverse-core.md provides detailed instructions for using dynamic analysis tools like Frida and various debuggers (IDA, Ghidra, radare2) for runtime artifact inspection.

xray