The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions in SKILL.md (Step 6) mandate the generation and execution of a Python script (/tmp/assemble_glossary.py) via the Bash tool. This script is used to parse, sort, and merge temporary files into the final glossary. Runtime generation and execution of scripts is a powerful capability that increases the attack surface if the instruction-following agent is manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary markdown files from the /docs directory to identify concepts and understand course context. This content serves as untrusted input that could contain malicious instructions.
Ingestion points: Reads docs/learning-graph/02-concept-list-v1.md, docs/course-description.md, and all files matching /docs/**/*.md.
Boundary markers: The skill lacks robust boundary markers or explicit instructions for the sub-agent to ignore embedded commands within the ingested concept lists or course descriptions.
Capability inventory: The skill has access to the Bash tool (for script execution), Read, and Write tools.
Sanitization: There is no evidence of sanitization, filtering, or validation of the input markdown content before it is interpolated into prompts for the definition generation agent.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform system-level tasks, specifically running Python to manage file operations and sorting. While intended for utility, this provides a mechanism for broader command execution if the agent's logic is subverted.

glossary-generator