The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to download and execute scripts from external community GitHub repositories that are not associated with trusted organizations.
Evidence: In references/patch-tuesday-workflow.md, it suggests cloning https://github.com/JaschaUrbach/msu-downloader.git and running the resulting msu-downloader.py script.
Evidence: Mention of using PatchExtract.ps1 (a community PowerShell script) to automate the extraction of multi-layer patch files.
Evidence: Instructions to clone https://github.com/joxeankoret/diaphora.git and execute it as a script within IDA Pro.
[EXTERNAL_DOWNLOADS]: The skill facilitates the acquisition of various external resources, including software patches, symbol files, and security tools.
Evidence: Fetches binary patches (MSU/MSP) from the Microsoft Update Catalog and Linux packages from official distribution repositories using apt and dnf.
Evidence: Downloads debug symbols from the Microsoft symbol server (msdl.microsoft.com/download/symbols).
Evidence: References the installation of the BinDiff tool from Google's official GitHub releases.
[COMMAND_EXECUTION]: The skill utilizes numerous system utilities and command-line tools to extract and process binary updates.
Evidence: Use of Windows utilities like expand.exe, dism.exe, and msiexec.exe for patch management.
Evidence: Use of Linux utilities like dpkg-deb, rpm2cpio, and cpio for extracting package contents.
[INDIRECT_PROMPT_INJECTION]: The skill processes complex external data that acts as a potential injection surface, though mitigated by mandatory confirmation gates.
Ingestion points: Untrusted binary patch files (.exe, .sys, .dll, .so), symbol files (.pdb, .dbgsym), and the output generated by automated diffing tools like ghidriff or BinDiff (specified in SKILL.md and references/diff-tools-comparison.md).
Boundary markers: Includes a "Mandatory confirmation gate" in SKILL.md requiring target identification, authorization check, and command explanation before execution.
Capability inventory: Comprehensive shell execution capabilities across all scripts, including file extraction (expand, dism), network downloads (git, apt, wget), and tool installation (pip, dpkg).
Sanitization: Lacks explicit sanitization or escaping instructions for external content when it is interpolated or summarized by an LLM for root cause analysis.
[DYNAMIC_EXECUTION]: The workflow involves the creation and execution of new code based on identified vulnerabilities.
Evidence: references/root-cause-and-poc.md provides templates and logic for developing Proof-of-Concept (PoC) code to trigger memory corruption (UAF, OOB, Race Conditions) and verify vulnerability persistence in unpatched versions.

patch-diff-exploit