The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax in SKILL.md to execute git log --oneline -5 2>/dev/null when the skill is loaded. This command retrieves the project's recent commit history to provide context for new commits. It is a read-only shell operation relevant to the skill's purpose and does not access sensitive credentials or perform network exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the project's git history, which is considered untrusted external input as it originates from potentially multiple contributors.
Ingestion points: The output of git log is displayed directly in the skill's context in SKILL.md via dynamic injection.
Boundary markers: There are no specific delimiters or instructions for the agent to ignore instructions that might be embedded within historical commit messages.
Capability inventory: The skill instructions lead to the execution of git add and git commit shell commands.
Sanitization: The skill does not sanitize or filter the content of the commit history before presenting it to the agent. This presents a surface where a malicious commit message in the repository history could attempt to influence the agent's current task, although the impact is restricted to the git workflow.

git-commit