The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends installing the optional 'pi' tool using curl -fsSL https://pi.dev/install.sh | sh, which executes a remote script directly in the shell without prior inspection.
[REMOTE_CODE_EXECUTION]: The script scripts/prepare-luna-catalog.sh dynamically generates and executes a Python script using a heredoc to process and modify JSON model catalog data.
[COMMAND_EXECUTION]: The scripts/bootstrap.sh and scripts/prepare-luna-catalog.sh scripts perform automated file system modifications, create directories, change file permissions (chmod +x), and execute other local shell scripts based on the project path.
[PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface where subagents like luna_scout and luna_worker ingest untrusted data from the codebase and environment while possessing workspace write capabilities.
Ingestion points: luna_scout reads the codebase, dependencies, and logs; luna_tester reads test logs.
Boundary markers: Defined in AGENTS.md and subagent instructions, advising agents on scope and escalation.
Capability inventory: luna_worker and luna_tester have workspace-write permissions; the main Sol agent can execute shell commands.
Sanitization: No explicit technical sanitization or escaping of ingested data is implemented in the configuration files.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs several Node.js packages including @openai/codex, @anthropic-ai/claude-code, and @earendil-works/pi-coding-agent from the NPM registry.
[SAFE]: The skill explicitly enforces security best practices by requiring API keys to be stored in environment variables and automatically adding .env files to .gitignore to prevent credential exposure.

sol-luna-setup