The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted project revisions and metadata to reproduce behavior changes, establishing an indirect prompt injection attack surface.
Ingestion points: Fetches external project revisions via git fetch and metadata via the collect_ty_ecosystem_run_metadata.py script.
Boundary markers: None identified; there are no explicit instructions to the agent to ignore or delimit embedded instructions in the analyzed data.
Capability inventory: The skill has significant capabilities including filesystem writes (mkdir, cp), software compilation (cargo build), package installation (uv run), and execution of dynamic shell commands.
Sanitization: No evidence of sanitization or validation for the project-specific commands or the content being minimized.
[COMMAND_EXECUTION]: The workflow requires executing shell commands to build the 'ty' binary using cargo build and subsequently running the generated binaries. It also executes dynamic commands output by the setup_primer_project.py helper script.
[EXTERNAL_DOWNLOADS]: The skill downloads the mypy-primer utility from a third-party GitHub repository (https://github.com/hauntsaninja/mypy_primer) using uv run. It also implicitly downloads Rust crates during the cargo build process.

minimizing-ty-ecosystem-changes