The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Python's subprocess module to interface with the local Git environment. Specifically, scripts/diff_surgeon.py executes git diff to retrieve changes for noise analysis. This execution is performed using a list of arguments without invoking a shell, which is a secure practice that prevents shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill operates by analyzing external and potentially untrusted data such as source code, git diffs, and project plans.
Ingestion points: Data enters the context through agents/karpathy-reviewer.md (which reads git diffs) and scripts/assumption_linter.py (which processes text proposals).
Boundary markers: No specific delimiters or safety instructions are used to wrap the ingested content.
Capability inventory: The skill has the capability to read any file the user provides and execute local Git commands via subprocesses.
Sanitization: The input is processed as raw text without sanitization.
While the Python analysis tools are regex-based and not susceptible to instruction overrides, the AI sub-agent (karpathy-reviewer) that reviews the diffs and tool outputs could potentially be influenced by instructions hidden within the code comments or documentation it is tasked with reviewing.

karpathy-coder