The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates documents containing data from the environment (git branch names, commit history, file status) and user-provided notes. These documents are intended for consumption by other developers or AI agents. If an attacker can influence the environment (e.g., via a malicious file name or commit message), they could inject instructions into the generated document.
Ingestion points: User-supplied NOTES and AGENT_TASK arguments, and outputs from git branch, git log, and git status (SKILL.md).
Boundary markers: Uses standard Markdown headers and list structures to separate content sections.
Capability inventory: Local file system operations (creating directories and writing Markdown files).
Sanitization: None observed; the skill performs direct string interpolation into the template.

handover