The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using python3 to run internal helper scripts (agent_runtime.py, codex_mcp_policy.py) and the codex CLI for configuration management tasks such as codex plugin list, codex mcp add/remove, and auth login.
[DYNAMIC_EXECUTION]: The instructions allow the agent to install a persistent Logfire CLI using a bundled helper script (agent_runtime.py --install) when the runtime is not present, facilitating environment setup and dependency management via uv.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret JSON output from CLI tools to determine the state of authentication, projects, and MCP connections.
Ingestion points: Instructions in references/connect.md and references/status.md require the agent to parse the output of commands like auth status and agent codex status.
Boundary markers: The skill specifically requests --output json to ensure the agent receives structured data rather than unpredictable natural language.
Capability inventory: The skill uses subprocess calls to python3 and the codex CLI to perform configuration writes and state inspections.
Sanitization: The skill relies on the standard JSON output formatting of the vendor-provided CLI tools.

logfire-codex-connection