The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill is designed to modify ~/.claude/settings.json, specifically the permissions.allow section. This allows it to grant persistent bash execution capabilities to other skills.
[COMMAND_EXECUTION] (HIGH): The /skill-permissions inject command explicitly states it can batch modify all SKILL.md files in the user's directory, which is a high-risk operation for file integrity and potential persistence of malicious instructions.
[PROMPT_INJECTION] (MEDIUM): By scanning other SKILL.md files for bash patterns to create allow-rules, this skill is vulnerable to Indirect Prompt Injection. A malicious skill could include a specifically crafted bash-like pattern in its markdown that this skill might parse and add to the global settings.json allowlist, bypassing intended security boundaries.
[COMMAND_EXECUTION] (LOW): The skill uses find and cat commands on the ~/.claude/ directory. While these are used for its primary purpose, accessing hidden configuration directories is a sensitive operation.

skill-permissions