The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data access were identified in the skill instructions or script.
[COMMAND_EXECUTION]: The skill executes a bundled local shell script (scripts/count.sh) using the run_skill_script tool to process diff metadata.
Evidence: The script uses standard Unix utilities (grep, printf) to count additions and deletions and handles input safely via standard input.
[PROMPT_INJECTION]: The skill ingests untrusted data from user-provided git diffs, which creates a surface for indirect prompt injection.
Ingestion points: Diff content provided by the user in the prompt or read via fs_read from local .diff or .patch files.
Boundary markers: Absent; there are no specific instructions to the agent to treat diff content as untrusted or to use delimiters to prevent instruction following within the diff.
Capability inventory: The skill has access to read local files (fs_read) and execute its bundled statistics script.
Sanitization: None; the agent is instructed to read the raw diff content directly to formulate its response.

git-diff-summariser