The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to interact with the local filesystem to manage the agent's environment. This includes listing directories in config paths (e.g., ~/.gemini/config/skills/), reading and modifying configuration files (skills.json, GEMINI.md), and proposing the deletion of redundant skill directories. These operations are restricted to the agent's own configuration scope and are central to the skill's primary purpose.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it audits external data from the disk inventory to influence its organization strategy.
Ingestion points: Skill directory names and metadata found in ~/.gemini/config/skills/ and workspace .agents/skills/, as well as the contents of skills.json.
Boundary markers: The instructions do not specify explicit delimiters for the data read from disk, but they instruct the agent to present findings as a "Pre-Execution Proposal" to the user.
Capability inventory: The skill uses the agent's ability to read/write files and delete directories to perform its organizational tasks.
Sanitization: There is no automated sanitization of filenames or directory contents; however, the skill mandates a user review phase where all proposed actions and metrics must be explicitly confirmed by the user before execution.

organise-agent-skills