The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves reading README files and source code from arbitrary, user-provided GitHub repositories to generate documentation and usage patterns.
Ingestion points: Step 3 fetches README.md and entry-point source files from remote repositories.
Boundary markers: Absent; the instructions do not include specific delimiters or warnings to ignore embedded natural language instructions within the processed source data.
Capability inventory: The skill can execute shell commands (python3, gh, npx), write files to the local system (Step 5), and perform network operations.
Sanitization: Content is summarized by the LLM in Step 4, which may inadvertently adopt malicious instructions found in the repository's documentation.
[COMMAND_EXECUTION]: The workflow extensively utilizes shell commands to execute local Python scripts and CLI tools for metadata resolution and validation.
Evidence: Found in Steps 1, 3, 4, 5, and 6 using python3, gh api, and npx.
Detail: User-provided inputs such as package names are regex-validated ([@a-zA-Z0-9._/-]+) to prevent shell metacharacter injection before being passed to subprocesses.
[EXTERNAL_DOWNLOADS]: The skill connects to external registries and GitHub to resolve and download package source code.
Detail: Queries registry.npmjs.org, pypi.org, and crates.io for package metadata. It uses the GitHub CLI (gh api) to retrieve repository contents and tags.
Note: These downloads target well-known technology services and are used for the skill's primary intended purpose.
[DATA_EXFILTRATION]: While not exfiltrating local data, the skill retrieves remote data and writes it to the local file system in Step 5.
Evidence: Compiled SKILL.md and metadata.json files are written to the {skill_package} directory.

skf-quick-skill