The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python scripts (folder_organization.py, export_portable.py) that perform local file system operations using standard libraries. These scripts implement safety features such as read-only inventory by default, directory traversal prevention, and collision detection to ensure no files are accidentally overwritten.
[INDIRECT_PROMPT_INJECTION]: The skill processes local file content to identify organizational markers, which presents a surface for indirect prompt injection from untrusted files.
Ingestion points: The scripts/folder_organization.py script reads the names, metadata, and content prefixes of files in a user-defined root directory to identify semantic markers (e.g., CLUE, OUTDATED).
Boundary markers: The instructions mandate a multi-signal verification process (checking rules, manifests, and content) and require a human-reviewed dry-run report before any file mutations are performed.
Capability inventory: The skill possesses capabilities for reading file metadata, calculating SHA-256 hashes, and performing file moves, renames, and archival operations within a local root.
Sanitization: The folder_organization.py script identifies and ignores markers located within markdown code fences or indented code blocks to prevent unintended interpretation of example text as control signals.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill incorporates a strict secrets policy to prevent the exposure of sensitive data. It uses a list of protected file patterns (e.g., .env, private keys) that are excluded from content reading and hashing. If a potential secret is discovered incidentally during a content scan, the script is designed to immediately stop processing and report only a signal ID, ensuring the sensitive value is not included in logs or reports.

folder-organization