The Agent Skills Directory

[COMMAND_EXECUTION]: Several scripts, including scripts/sample-code-quality.py and scripts/compute-code-rating.py, utilize subprocess.run to call the Git CLI.
Evidence: Functions such as git_log_sample, git_show_diff, and git_show_file execute shell commands to retrieve repository history and diffs.
Context: This behavior is necessary and expected for a tool designed to analyze git data. The commands are invoked using list-based arguments to prevent injection, and the inputs are scoped to local repository paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external CSV files and local git repository histories to generate markdown reports. This represents a potential indirect prompt injection surface.
Ingestion points: Data enters the system via raw-commits.csv, mr-acceptances.csv, and through file reads in repository paths defined in repo_roots (scripts/extract-contribution-profile.py, scripts/sample-code-quality.py).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands were observed in the ingestion logic.
Capability inventory: The skill can perform shell execution via subprocess.run and write files to the local system (reports and JSON data).
Sanitization: The scripts perform standard CSV parsing and AST analysis but do not include specific sanitization against LLM-targeted injection patterns in git commit messages or code comments.
Assessment: Given the analytical nature of the skill, this is a low-risk surface common to data processing tools.

dev-contribution-quality-analysis