The Agent Skills Directory

[SAFE]: The skill instructions are focused entirely on improving Git commit quality and granularity. No attempts at prompt injection, obfuscation, or persistence were detected.
[COMMAND_EXECUTION]: The skill uses local git commands (git status, git add) which are necessary for its stated purpose of managing commit granularity. No unauthorized or dangerous command execution was found.
[DATA_EXFILTRATION]: There are no network calls or attempts to access sensitive system directories (like .ssh or .aws). The skill only interacts with the local repository's status and diffs.
[INDIRECT_PROMPT_INJECTION]: The skill has a minimal attack surface for indirect prompt injection as it processes untrusted data from git status and diffs. However, its capabilities are strictly limited to local Git staging, and it includes explicit instructions to ask the user for clarification if commit boundaries are unclear.
Ingestion points: Processes output from git status and file diffs.
Boundary markers: Instructs the agent to consult the user if the split boundary is unclear.
Capability inventory: Limited to staging files using git add and classifying changes by intent.
Sanitization: None required for the limited scope of local staging operations.

commit-granularity