From cc1de21092eda10667dabe2122a8b75b66800aa4 Mon Sep 17 00:00:00 2001 From: Gauransh Mathur Date: Sun, 4 Oct 2026 18:59:37 +0800 Subject: [PATCH] fix(advise): use both advisors inside a subagent when it can start one The skill and the README said a subagent cannot start another subagent, so the skill always skipped the Fable advisor there. Claude Code lets a subagent start its own subagents up to three layers deep, and withholds the Agent tool only at that limit or when the definition leaves it out. Inside a subagent the skill now dispatches the Fable advisor when it has the Agent tool, and runs Codex alone only when it does not. It still never uses AskUserQuestion there, and hands any disagreement back to the parent. Closes #28 --- README.md | 8 +++++--- skills/advise/SKILL.md | 14 ++++++++------ 2 files changed, 13 insertions(+), 9 deletions(-) diff --git a/README.md b/README.md index 9e5864b..34c9db3 100644 --- a/README.md +++ b/README.md @@ -91,9 +91,11 @@ It never runs on Claude's own initiative. There are three ways to start it: changes and include its verdict in your report. ``` - A subagent cannot start another subagent or ask you a question, so inside - one the skill runs Codex alone as the second model and hands any - disagreement back to the main conversation instead of asking you. + A subagent cannot ask you a question, so inside one the skill hands any + disagreement back to the main conversation instead of asking you. It still + uses both advisors when the subagent is allowed to start one of its own. + When it is not, because its definition leaves out the `Agent` tool or it is + at Claude Code's nesting limit, the skill runs Codex alone. ## Data handling diff --git a/skills/advise/SKILL.md b/skills/advise/SKILL.md index 063c1ea..01c5e5a 100644 --- a/skills/advise/SKILL.md +++ b/skills/advise/SKILL.md @@ -42,14 +42,16 @@ Never run it on every edit. Two runs per task is the whole budget. ### Inside a subagent When you are a subagent that was told to run this as a QA check, run the -`done` checkpoint once, before you report back. A subagent cannot start -another subagent or ask the user, so: +`done` checkpoint once, before you report back. Two things differ there: -- Skip the Fable advisor and run Codex alone. Report the Fable advisor as +- A subagent has the `Agent` tool only when its definition allows it and it + is not at the nesting limit. If you have it, dispatch the Fable advisor as + usual. If you do not, run Codex alone and report the Fable advisor as skipped, not as failed. -- Do not use `AskUserQuestion`. When Codex disagrees with your work on - anything that changes the verdict, put both positions in your report to the - parent and let it decide. +- A subagent cannot ask the user, so do not use `AskUserQuestion`. When the + advisors disagree with your work, or with each other, on anything that + changes the verdict, put the positions in your report to the parent and let + it decide. ## Step 1: Write one shared brief