The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/scope_creep.py executes local git commands using subprocess.run to retrieve diff data. The command construction uses a fixed list of arguments rather than a shell, mitigating command injection risks, and is a legitimate requirement for the skill's primary function.
[PROMPT_INJECTION]: The skill processes git diff output which can contain content from untrusted sources, such as third-party pull requests. This represents a potential surface for indirect prompt injection if the diff content contains malicious natural language instructions.
Ingestion points: Repository data and diffs are ingested via the read_diff function in scripts/scope_creep.py.
Boundary markers: The instructions do not define specific delimiters or encapsulation techniques to separate untrusted diff content from the agent's system instructions.
Capability inventory: The skill possesses the ability to execute local git commands and perform file system reads through the provided Python script.
Sanitization: The Python script parses the unified diff structure (headers and hunks) but does not sanitize or filter the actual source code or comments for potential agent instructions.

scope-creep-detector