The Agent Skills Directory

[PERSISTENCE]: The skill instructs the agent to modify project-wide configuration files (e.g., CLAUDE.md, AGENTS.md, .cursor/rules/) to inject mandatory instructions for future sessions. This ensures that the agent always prioritizes using the godot-prompter toolset in every session.
[DATA_EXPOSURE]: The skill requires the agent to read from and write to a directory in the user's home path (~/.godot-prompter/state/<hash>.json). Accessing and modifying the filesystem outside of the project root, particularly within the home directory, is a significant security risk used to track cross-project behavior.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by reading and modifying agent instruction files that are processed at the start of every session.
Ingestion points: CLAUDE.md, CLAUDE.local.md, .claude/CLAUDE.md, AGENTS.md, GEMINI.md, .github/copilot-instructions.md, and the .claude/rules/ and .cursor/rules/ directories.
Boundary markers: Absent for the injected instruction block.
Capability inventory: Capabilities to write to project configuration files and the global home directory.
Sanitization: Absent; the skill instructs the agent to write a specific, hardcoded instruction block into the files.
[TIME_DELAYED_CONDITIONAL]: The execution of the instruction injection is gated by a check for previous user refusals stored in the hidden state file (~/.godot-prompter/state/<hash>.json).

godot-brainstorming