fable-sonnet — Claude Code skill: what it does, how to install
Enforces a disciplined execution loop on Claude Sonnet with failable step checks, parallel delegation, and skeptical self-critique.
npx skills add mrtooher/fable-mode --skill fable-sonnet -a claude-codeWhat fable-sonnet does
The skill applies structured task execution by spawning a Claude Sonnet subagent. Instead of tackling a prompt in a single pass, the agent writes out a numbered stage plan where each phase must produce an inspectable output. It allows parallel agents for independent parts, capping delegation at a single level to prevent runaway costs.
Its main focus is enforcing checks that can actually fail. Sonnet frequently substitutes automated verification with superficial reading. To counter this, the procedure strictly requires concrete proof—diffing outputs against specifications, inspecting created files, or running command-line tests. The workflow permits at most two complete replans; a third indicates unclear scope and prompts the agent to halt.
Before output delivery, the process demands a critical review and passes deliverables through the double-check skill. Without the Agent tool available in your environment, the routine executes inline within the active chat without pinning Sonnet. If an assignment exceeds Sonnet's reasoning boundaries, the guidance directs the agent to halt and propose upgrading to fable-opus.
What it helps with
- break complex engineering jobs into verifiable stages
- stop the model from faking validation with visual review
- run parallel subagents without nesting overhead
- catch subtle flaws through skeptical pre-delivery critique
Example prompts for the agent
- Use fable-sonnet to plan and build the caching layer with tests for each phase.
- Run fable on sonnet to refactor the data parser and verify output shapes at every step.
How to install fable-sonnet
1Skills CLI (Vercel)
npx skills add mrtooher/fable-mode --skill fable-sonnet -a claude-code2Manually: put the files in place
~/.claude/skills/fable-sonnet/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/fable-sonnet/
1Skills CLI (Vercel)
npx skills add mrtooher/fable-mode --skill fable-sonnet -a codex2Manually: put the files in place
.agents/skills/fable-sonnet/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add mrtooher/fable-mode --skill fable-sonnet -a cursor2Manually: put the files in place
.cursor/skills/fable-sonnet/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install mrtooher/fable-mode fable-sonnet2Manually: put the files in place
.github/skills/fable-sonnet/SKILL.md1Skills CLI (Vercel)
npx skills add mrtooher/fable-mode --skill fable-sonnet -a gemini-cli2Platform CLI
gemini skills install https://github.com/mrtooher/fable-mode --path fable-sonnetManually: put the files in place
.devin/skills/fable-sonnet/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add mrtooher/fable-mode --skill fable-sonnet -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about fable-sonnet
How to install fable-sonnet in Claude Code?
Run npx skills add mrtooher/fable-mode --skill fable-sonnet -a claude-code in your terminal.
What does fable-sonnet do?
It spins up a Claude Sonnet worker following a four-part workflow: staging outputs, running tests that can fail, performing parallel subtasks, and completing an unsparing final review.
What happens if subagent spawning is unavailable?
When the runtime lacks the Agent tool or lacks explicit spawn authorization, the discipline executes inline inside the existing conversation without model pinning.
