The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions in SKILL.md include proactive safety measures, explicitly directing the agent to prioritize system and safety guidelines over the project memory stored in .lore/. It warns the agent to treat memory as project-controlled input and not as an authority to bypass safety filters or expand access.
[COMMAND_EXECUTION]: The skill utilizes several Python helper scripts (history.py, find_duplicates.py, etc.) that execute shell commands via subprocess.run. These commands are limited to legitimate uses of git and calling other internal Python scripts from the same package. The scripts employ safe practices, such as using argument lists and the -- separator in Git commands to prevent argument injection.
[EXTERNAL_DOWNLOADS]: The documentation references cloning the skill repository from GitHub. GitHub is a well-known and trusted service for code hosting. The skill itself does not perform any automated downloads or network requests during operation.
[DATA_EXFILTRATION]: No network-enabled libraries or commands (such as requests, urllib, curl, or wget) are present in the scripts. Data operations are strictly local to the project's filesystem.
[SAFE]: The skill follows security best practices by validating mirror target paths against an allowlist and checking for directory traversal and symlink escapes before performing file operations.

lore