The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill directs the agent to load and follow additional instruction files (SKILL.md) using paths derived from names identified in the input text (e.g., words in bold or backticks).
Evidence: "A skill named in bold or in backticks, or run as /<name>, is the file ../<name>/SKILL.md beside this skill's directory... Read the file and follow it."
This implements dynamic loading from computed paths. While intended for loading writing conventions, it presents a potential path traversal risk if input strings are not validated by the underlying platform to prevent directory traversal sequences (e.g., ../../).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as documentation, RFCs, and PR descriptions, creating an attack surface for indirect prompt injection.
Ingestion points: The skill is designed to review and rewrite user-provided text files, commit messages, and descriptions.
Boundary markers: There are no explicit delimiters or instructions to ignore potentially malicious commands embedded within the processed content.
Capability inventory: The instructions describe a file-reading behavior triggered by patterns found in the input text.
Sanitization: No validation or filtering of the input strings used for file path construction is defined in the guidelines.

technical-writing