The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is configured with the Bash, Write, and Edit tools, which it uses to programmatically create new SKILL.md files in the .claude/skills/ directory. This allows the skill to persist new executable instructions that remain across sessions.
[COMMAND_EXECUTION]: The workflow incorporates Glob and Bash shell commands (e.g., find) to inspect and map the contents of the local skill library.
[PROMPT_INJECTION]: The 'Skill Creation Workflow' represents a significant indirect prompt injection surface. Because the skill creates new instruction sets based on identified 'gaps' or user requests, an attacker could potentially influence the creation of a malicious skill. Evidence Chain: 1) Ingestion points: The skill reads existing skill definitions from the filesystem via Glob. 2) Boundary markers: No delimiters or warnings are used to prevent the agent from obeying instructions embedded in the data it processes. 3) Capability inventory: The skill has the ability to use Bash, Write, and Edit to modify the agent's operating environment. 4) Sanitization: There is no evidence of sanitization or validation of the input used to generate the content of the new SKILL.md files.

team-builder