The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Git commit messages and file diffs. This creates an attack surface where a malicious commit could contain instructions designed to influence the agent's output.\n- Ingestion points: Git commit subjects (%s), bodies (%b), and file diffs retrieved via git log and git show as described in SKILL.md.\n- Boundary markers: The skill lacks instructions to wrap commit data in delimiters or provide specific warnings to the AI to ignore embedded instructions within the Git history.\n- Capability inventory: The agent executes shell commands (git) and reads repository file contents to generate the summary.\n- Sanitization: No specific sanitization, escaping, or filtering of commit content is mentioned in the logic to prevent instruction interpretation.\n- [DYNAMIC_EXECUTION]: The skill constructs and executes shell commands (git log, git show, git rev-parse) using parameters derived from user input, such as project paths and dates. While the skill instructs the agent to validate paths and avoid fuzzy matching, there is a residual risk of command injection if the execution environment does not properly escape these inputs before passing them to the shell.

daily-work-summary