The Agent Skills Directory

[SAFE]: The skill implements strong security practices in its core script, init_project_memory.py. It uses a resolve_inside function that ensures all file operations are restricted to the specified project root, effectively preventing path traversal attacks.
[SAFE]: The script includes a dedicated check (reject_symlink_components) that recursively validates each component of a path to ensure it does not involve symlinks, protecting against symlink-based file overwrites outside the intended directory.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to create and update instructions for future AI agents in AGENTS.md and associated memory files. This creates an attack surface for indirect prompt injection. However, the risk is mitigated by the following: \n- Ingestion points: The script reads the project root path and existing contents of the agent instruction file (e.g., AGENTS.md). \n- Boundary markers: Managed instructions are strictly delimited by comment markers ( and ). \n- Capability inventory: The script is limited to file reading, writing, and directory creation within a validated scope. It does not perform network operations or execute external commands. \n- Sanitization: Path validation ensures operations remain within the target project root.
[SAFE]: The skill's instructions in SKILL.md and the generated README.md explicitly warn against storing secrets, credentials, tokens, or private keys in the memory files, promoting secure data handling practices.
[SAFE]: The README.md provides an installation command fetching the skill from the author's own repository on GitHub.

memory-init