Claude

Skill zachmayer/skills/.claude/skills/claude

Sends a prompt to a fresh background Claude sub-agent for an outside read with full Claude Code tools. Use when the user says "ask Claude", "Claude sub-agent", "ask Fable", "ask Opus", "ask another Claude", "fresh Claude", "Claude second opinion", or wants a design/taste call, prose critique, or adversarial self-review of Claude Code output. Do NOT use for tasks Claude can answer directly, or in claude.ai chat — sub-agents run only in Claude Code and Cowork.From its SKILL.md

Install

npx -y skills add zachmayer/skills --skill claude

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.7 KB, as published. Nobody here has run it

Verify installed

The Agent tool is built into Claude Code. No CLI to run.

Preferences

subagent_type: general-purpose (full tool access; alternatives include Explore, Plan, code-simplifier, etc. — see the Agent tool's subagent types)
model: fable — the default: much smarter, worth it for outside reads. Fable is the more global model — strongest on open-ended problems where the framing is part of the question. opus is the cheaper, more local alternative: a bit more detail-oriented, better at working hard through well-defined but difficult problems.
context: fresh — the sub-agent sees only what's in prompt.

Always background

Claude sub-agents typically take a few minutes but can run for half an hour on hard problems. Set run_in_background: true on the Agent call. Continue with other work; surface results when the completion notification fires.

Send a prompt from a file

ts=$(date +%Y%m%d_%H%M%S)_$$
# Write prompt to ~/claude/scratch/prompt_claude_${ts}.md (Write tool), then:

Agent(
  subagent_type: "general-purpose",
  model: "fable",
  description: "Discussion partner — Claude",
  prompt: "Read your task at ~/claude/scratch/prompt_claude_<TS>.md and respond.",
  run_in_background: true
)

Substitute <TS> with the literal value of ts. The Agent tool isn't bash — ${ts} doesn't expand inside the prompt string.

Strengths and weaknesses

Design/taste calls, prose artifacts, adversarial self-review of Claude Code output, subtle state/concurrency.

Watch for family blind spots (the sub-agent is a Claude too — it shares biases with the parent, doubly so when both run the same model) and yes-manning (especially when called from Claude Code — Claude sub-agents tend to agree with prior framing). Pair with codex or gemini when independence matters; ask for the strongest objections explicitly.

Troubleshooting

Sub-agent returns immediately with no work done → prompt was too thin; sub-agents see zero conversation context.
Sub-agent reports an unexpected model → pass the model from Preferences explicitly.
"I cannot access that file" → the sub-agent didn't expand ~; give it the absolute path instead.