JoaquinCampo Advisor
Use this skill to reproduce the Claude Advisor workflow in Codex. The advisor must be a separate model call so it can judge the main agent's work from the outside.
Dispatch Requirement
Always run the advisor with:
model: "gpt-5.6-sol"reasoning_effort: "xhigh"- system prompt: the exact contents of
references/advisor_system_prompt.md - user message: the full available transcript of the agent's task, tool calls, tool results, failed attempts, current plan, and self-checks
Do not answer inline as the advisor when a subagent or separate model-call mechanism is available. If the runtime cannot spawn a separate gpt-5.6-sol call with xhigh reasoning, say that exact behavior is unavailable and ask for the runtime to be upgraded rather than silently using an older model family.
Transcript Construction
Pass the richest transcript the environment can provide:
- original user request and constraints
- every relevant tool call and result
- errors, retries, dead ends, and corrections
- current files or diffs under consideration
- previous advisor advice, if any
- the agent's current question, if there is one
If there is no explicit question, ask the advisor to follow the recovered prompt exactly: infer where the agent is and give direct actionable advice.
Prompt Files
references/advisor_system_prompt.md: exact advisor system prompt to use for the separate model call.references/advisor_prompt.md: original recovered note that this skill was derived from.
Completion criterion: the returned advice came from a separate gpt-5.6-sol xhigh advisor call using references/advisor_system_prompt.md, or the response clearly states why exact advisor behavior could not be run.