The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/diff_surgeon.py executes the git diff command using the Python subprocess module to analyze file changes. The implementation uses list-based arguments which is a secure method for executing local binaries.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted user content such as source code and project plans, creating an attack surface for indirect prompt injection. 1. Ingestion points: scripts/complexity_checker.py, scripts/diff_surgeon.py, scripts/assumption_linter.py, and scripts/goal_verifier.py read file contents and git output. 2. Boundary markers: No delimiters or explicit instructions are used to isolate untrusted data. 3. Capability inventory: The tools are limited to reading local files and executing the git binary. 4. Sanitization: No content filtering is applied.
[METADATA_POISONING]: The SKILL.md frontmatter lists an author that does not match the provided skill author, representing a metadata inconsistency.

karpathy-coder