The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local command execution to automate repository management. This includes running bash and python scripts for peer discovery (find-peers.sh), linting (lint-skill.py), and version management (bump-version.sh). It also uses rm -rf to clean up temporary build artifacts generated during the skill creation process and utilizes the ! command syntax in SKILL.md to execute shell commands at load time for environment checks.
[PROMPT_INJECTION]: There is an attack surface for indirect prompt injection as the skill collects user input through an interview process and interpolates these strings into a handoff template for the skill-creator plugin. 1. Ingestion points: User survey responses in Phase 1 and Phase 2b of the authoring workflow. 2. Boundary markers: No explicit delimiters or 'ignore embedded instructions' warnings are used in the handoff preamble to the skill-creator tool. 3. Capability inventory: The skill has the ability to execute shell commands, delete directories, and write files to the repository through the orchestration layer. 4. Sanitization: No sanitization or escaping of user input was observed before it is interpolated into the prompt templates for the downstream agent.

toolkit-skill-author