The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection when handling complex cases that involve a sub-agent.\n
Ingestion points: The skill reads git diff and git diff --stat output from the local repository (SKILL.md), which may contain malicious instructions embedded in file comments or code.\n
Boundary markers: The prompt template for the commit-manager sub-agent lacks robust delimiters or explicit instructions to treat the diff content as data only, increasing the risk that the sub-agent follows instructions found within the diff.\n
Capability inventory: The skill uses the Task tool to invoke sub-agents and executes shell commands like git commit and git add.\n
Sanitization: There is no evidence of filtering or escaping the diff output before it is interpolated into the prompt for the sub-agent.\n- [COMMAND_EXECUTION]: The skill performs shell command execution to interact with the version control system, specifically using git diff, git add, and git commit. These operations are consistent with the skill's purpose but involve direct interaction with the host environment's file system and repository state.

commit-diffs