🎯Skills3
Internal helper contract used by the `codex:codex-rescue` subagent to call the codex-companion runtime from Claude Code. Forwards the user's request through exactly one `task` invocation (with optional `--effort`, `--model`/`spark`, `--resume-last`, `--fresh`, `--write`), strips Claude-side flags, and returns Codex stdout verbatim — no repo inspection or follow-up work on the Claude side.
Internal Codex-plugin skill for composing GPT-5.4 / Codex prompts used by `codex:codex-rescue` — compact XML-structured blocks (`<task>`, `<structured_output_contract>`, `<default_follow_through_policy>`, `<verification_loop>`, `<grounding_rules>`, `<action_safety>`), one clear job per run, and tightening the prompt before raising reasoning effort.
Internal guidance for the Codex plugin on how to present Codex CLI output — preserve verdict/summary/findings structure, order by severity, keep file/line references exact, and never silently patch after a failed run.