The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains directive instructions using 'YOU MUST', 'Don't rationalize', and 'Critical Rules' which attempt to override the agent's default decision-making process and enforce rigid compliance with external skill files.
[COMMAND_EXECUTION]: The skill-run script implements a generic execution primitive that allows the agent to run any executable file residing within the skills directory tree by using the exec command.
[COMMAND_EXECUTION]: The find-skills script utilizes shell utilities like find and grep to search through skill files on the local filesystem based on potentially unvalidated patterns provided to the script.
[PROMPT_INJECTION]: Analysis of the Indirect Prompt Injection surface within the skill framework:
Ingestion points: Data enters the system context when the agent reads SKILL.md files discovered by the find-skills tool or specified via file paths.
Boundary markers: There are no identified boundary markers or instructions to ignore embedded commands; instructions explicitly command the agent to 'Follow it exactly' and read the 'ENTIRE file'.
Capability inventory: The skill includes tools for searching the filesystem (find-skills), executing scripts (skill-run), and managing persistent tasks (TodoWrite).
Sanitization: No validation, sanitization, or filtering of the content within the processed skill files is performed before the agent is instructed to act upon them.

Getting Started with Skills