The Agent Skills Directory

[COMMAND_EXECUTION]: Several Python scripts execute external shell commands via the subprocess module to manage project resources and skill installations. Evidence includes calls to git for repository management and pnpx for package searching and installation in scripts/official.py, scripts/resources.py, and scripts/global_setup.py.\n- [REMOTE_CODE_EXECUTION]: The skill facilitates the installation of third-party agent skills by downloading code directly from remote GitHub repositories and the skills.sh registry. The pnpx skills add and git clone commands integrated into the workflow allow for the execution of remote code within the agent's project environment.\n- [EXTERNAL_DOWNLOADS]: The skill interacts with external network resources to discover and fetch content. It queries https://www.skills.sh for skill metadata and clones repositories from https://github.com using the git CLI. It also uses the pnpx utility to download and install packages from the npm registry.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from the project being analyzed (source code, manifests, and documentation) and from the descriptions of external skills found during discovery. This creates a surface where malicious instructions could be embedded to influence the agent's behavior during the selection and installation process.\n
Ingestion points: Project files read in scripts/analysis.py and fetched skill instructions read in scripts/resources.py.\n
Boundary markers: The skill instructions explicitly warn the agent to review candidate content as material and not to execute bundled scripts during the discovery phase.\n
Capability inventory: The skill possesses capabilities for file system modification (creating symbolic links), command execution (git, pnpx), and network access.\n
Sanitization: Input validation using regular expressions is implemented for repository and skill names in scripts/official.py and scripts/resources.py.

boost