The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local project files that could contain malicious instructions designed to influence the behavior of the sanitizer or packager sub-agents.
Ingestion points: Arbitrary project files are read from SOURCE_PATH as described in the protocols for the opensource-forker, opensource-sanitizer, and opensource-packager agents in SKILL.md.
Boundary markers: The skill does not explicitly use delimiters or "ignore instructions" warnings when passing project content to sub-agents.
Capability inventory: The skill possesses the ability to read and write files, execute shell commands, and create public GitHub repositories using the gh CLI.
Sanitization: The workflow includes a mandatory opensource-sanitizer sub-agent designed to audit for secrets, PII, and dangerous files, providing a specific defensive layer against accidental data exposure.
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage a local staging environment and interact with GitHub's CLI tool.
Evidence: The protocol in SKILL.md uses mkdir -p to create staging areas, ls and cat to manage staging status, and gh repo create to publish the final project to GitHub.
[DYNAMIC_EXECUTION]: The skill generates and prepares executable scripts for the final project package.
Evidence: In Step 5 of the protocol in SKILL.md, the opensource-packager agent is instructed to generate a setup.sh bootstrap script and ensure it is marked as executable.

opensource-pipeline