The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts new-note.py and verify-notes.py utilize the subprocess module to execute git commands for repository state inspection, such as identifying changed files, reading commit history (git log), and resolving file paths via git ls-files. These commands use fixed arguments and do not involve shell interpolation of untrusted input.
[PERSISTENCE]: The skill includes an --install command that automates the setup of a git pre-commit hook. This hook is designed to automatically run the verify-notes.py script before every commit to ensure consistency. While this constitutes an automated execution mechanism, it is a documented core feature intended to enforce the skill's safety "gate" for developers.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by reading and displaying the contents of user-contributed markdown "notes" to the AI agent. If a repository contains maliciously crafted notes, they could potentially influence the agent's behavior during a session.
Ingestion points: verify-notes.py (via --find and --for-path flags) and new-note.py (during duplicate checks) read all .md files within the configured notes directory.
Boundary markers: None are explicitly used when displaying note content to the agent.
Capability inventory: The skill allows the agent to modify local repository documentation (AGENTS.md, CLAUDE.md) and git hooks.
Sanitization: The scripts perform regex-based parsing of note headers but treat the note body as plain text for display.
[DYNAMIC_EXECUTION]: new-note.py uses importlib.util to dynamically load its sibling script verify-notes.py. This is used for code reuse between the two local scripts and does not involve loading remote or untrusted code.

regression-notes