From 99bc9d48ad1c48bf015422cbadbb714d4ba78d7b Mon Sep 17 00:00:00 2001 From: Paul Bakaus Date: Thu, 2 Apr 2026 20:29:18 -0700 Subject: [PATCH] Revert critique sub-agent over-engineering, add Codex spawning guidance Reverts Codex's overly bureaucratic sub-agent delegation changes (permission prompts, "not fully compliant" disclaimers) back to clear, practical language. Uses RFC-style SHOULD for sub-agent delegation with named examples for both Claude Code (Agent tool) and Codex (natural language spawning). Co-Authored-By: Claude Opus 4.6 (1M context) --- .agents/skills/critique/SKILL.md | 4 +++- .claude/skills/critique/SKILL.md | 4 +++- .codex/skills/critique/SKILL.md | 4 +++- .cursor/skills/critique/SKILL.md | 4 +++- .gemini/skills/critique/SKILL.md | 4 +++- .kiro/skills/critique/SKILL.md | 4 +++- .opencode/skills/critique/SKILL.md | 4 +++- .pi/skills/critique/SKILL.md | 4 +++- .trae-cn/skills/critique/SKILL.md | 4 +++- .trae/skills/critique/SKILL.md | 4 +++- source/skills/critique/SKILL.md | 4 +++- 11 files changed, 33 insertions(+), 11 deletions(-) diff --git a/.agents/skills/critique/SKILL.md b/.agents/skills/critique/SKILL.md index 725ae3a9e..9e2bf6e4c 100644 --- a/.agents/skills/critique/SKILL.md +++ b/.agents/skills/critique/SKILL.md @@ -15,7 +15,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.claude/skills/critique/SKILL.md b/.claude/skills/critique/SKILL.md index e1b17b147..337d8ea6a 100644 --- a/.claude/skills/critique/SKILL.md +++ b/.claude/skills/critique/SKILL.md @@ -20,7 +20,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.codex/skills/critique/SKILL.md b/.codex/skills/critique/SKILL.md index dd7e4fe2a..52e2584a5 100644 --- a/.codex/skills/critique/SKILL.md +++ b/.codex/skills/critique/SKILL.md @@ -14,7 +14,9 @@ Invoke $impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.cursor/skills/critique/SKILL.md b/.cursor/skills/critique/SKILL.md index 7da87c278..6b3f0a8be 100644 --- a/.cursor/skills/critique/SKILL.md +++ b/.cursor/skills/critique/SKILL.md @@ -13,7 +13,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.gemini/skills/critique/SKILL.md b/.gemini/skills/critique/SKILL.md index 1d446d078..cea8e2224 100644 --- a/.gemini/skills/critique/SKILL.md +++ b/.gemini/skills/critique/SKILL.md @@ -13,7 +13,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.kiro/skills/critique/SKILL.md b/.kiro/skills/critique/SKILL.md index a273aee2c..e0180b5bf 100644 --- a/.kiro/skills/critique/SKILL.md +++ b/.kiro/skills/critique/SKILL.md @@ -13,7 +13,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.opencode/skills/critique/SKILL.md b/.opencode/skills/critique/SKILL.md index b992f3f24..dc5008b83 100644 --- a/.opencode/skills/critique/SKILL.md +++ b/.opencode/skills/critique/SKILL.md @@ -20,7 +20,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.pi/skills/critique/SKILL.md b/.pi/skills/critique/SKILL.md index 3b7a34a70..9d1c1142c 100644 --- a/.pi/skills/critique/SKILL.md +++ b/.pi/skills/critique/SKILL.md @@ -18,7 +18,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.trae-cn/skills/critique/SKILL.md b/.trae-cn/skills/critique/SKILL.md index 2c0a360b6..12edaf79b 100644 --- a/.trae-cn/skills/critique/SKILL.md +++ b/.trae-cn/skills/critique/SKILL.md @@ -15,7 +15,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/.trae/skills/critique/SKILL.md b/.trae/skills/critique/SKILL.md index 2c0a360b6..12edaf79b 100644 --- a/.trae/skills/critique/SKILL.md +++ b/.trae/skills/critique/SKILL.md @@ -15,7 +15,9 @@ Invoke /impeccable -- it contains design principles, anti-patterns, and the **Co Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state. diff --git a/source/skills/critique/SKILL.md b/source/skills/critique/SKILL.md index c694d489a..04bd9936a 100644 --- a/source/skills/critique/SKILL.md +++ b/source/skills/critique/SKILL.md @@ -20,7 +20,9 @@ Invoke {{command_prefix}}impeccable -- it contains design principles, anti-patte Launch two independent assessments. **Neither must see the other's output** to avoid bias. -If your system supports sub-agents (e.g., Claude Code's Agent tool), delegate each assessment to a separate agent. The sub-agents should return their findings as structured text -- do NOT output findings to the user yet. If sub-agents are not available, complete each assessment sequentially, writing findings to internal notes before proceeding. +You SHOULD delegate each assessment to a separate sub-agent for independence. Use your environment's agent spawning mechanism -- e.g., Claude Code's `Agent` tool, or Codex's subagent spawning ("spawn two agents in parallel"). Sub-agents should return their findings as structured text -- do NOT output findings to the user yet. + +If sub-agents are not available in the current environment, complete each assessment sequentially, writing findings to internal notes before proceeding. **Tab isolation**: When browser automation is available, each assessment MUST create its own new tab -- never reuse an existing tab, even if one is already open at the correct URL. This prevents the two assessments from interfering with each other's page state.