The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user instructions to modify local configuration files, which represents an indirect prompt injection surface.
Ingestion points: User requests concerning Zed configuration, themes, keybindings, and LSP settings.
Boundary markers: Not explicitly defined in the prompt instructions.
Capability inventory: The skill is authorized to read and edit configuration files (settings.json, keymap.json) and use bash tools for JSON manipulation (jq).
Sanitization: The skill references an external validation script (.codex/skills/zed-config/scripts/validate-config.sh) to ensure modified configuration files maintain valid JSON syntax.
[CREDENTIALS_SAFE]: The documentation explicitly instructs the agent that API keys for LLM providers should be stored in the OS keychain rather than the settings files. It provides examples using standard environment variable names (e.g., ANTHROPIC_API_KEY, OPENAI_API_KEY) without hardcoding actual secrets.
[EXTERNAL_DOWNLOADS]: The reference documentation contains links to official Zed documentation and well-known AI provider services (e.g., zed.dev, anthropic.com, openai.com, ollama.com, openrouter.ai). These are well-known services and are documented neutrally as sources for configuration parameters.
[COMMAND_EXECUTION]: The skill provides documentation for Zed's task system, which allows users to define commands that run in the integrated terminal. While this system executes commands, the skill's role is strictly limited to helping the user configure these tasks within the editor's expected workflow.

zed-config