Fable prompt
Translate a validated plan, decision, or brainstorm output into a Claude Fable 5 agent prompt — goal, constraints, feedback signal, and stop condition. Use when the user says "turn this into a Fable prompt", "generate the Fable prompt", "make this loop-ready", or has just finished a grilling/ideation/steelman session and wants to hand the result to an agent harness.From its SKILL.md
npx -y skills add atalariq/skills --skill fable-promptAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 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
2.8 KB, as published. Nobody here has run it
Fable Prompt
Translate a resolved plan into a prompt built for how Fable 5 actually performs well: assign a goal and a verification loop, not a one-shot instruction.
Only reach for this once the plan is resolved (no open decisions or unvalidated assumptions — run grilling or ideation first if not) and the task is big enough to warrant loop scaffolding. For a short/routine task or a five-minute fix, write a normal prompt instead.
Input
Whatever came out of a prior session: a grilling summary, an ideation verdict, a steelman/pre-mortem output, or the user's own plan in prose.
Process
1. Extract the goal
One sentence. A deliverable, not an activity. ("Ship a working MVP that does X" not "work on X".)
Completion criterion: the goal names a concrete artifact or end state, not a process.
2. Extract constraints
Pull from decisions already locked in the source material — tech stack, scope boundaries, things explicitly ruled out. Don't invent new ones.
Completion criterion: every constraint traces back to a decision the user already made, not an assumption you're adding.
3. Write the feedback signal
What can a verifier check without a human reading prose? Prefer: tests passing, visual diff against a reference, a rubric with checkable line items. Avoid vague criteria like "do it well" or "make sure it's good."
If the source material included a pre-mortem, turn the top 1-2 risks into things the feedback signal explicitly checks for.
Completion criterion: a separate agent or script could evaluate the signal without asking the user anything.
4. Write the stop condition
The exact state that ends the loop. Include a max-iteration or timeout fallback so the loop can't run forever on an ambiguous signal.
5. Assemble
Use this structure:
Goal: [one sentence]
Constraints:
- [constraint 1]
- [constraint 2]
Feedback signal:
- [checkable item 1]
- [checkable item 2]
Stop condition: [state that ends the loop, plus iteration/time cap]
Output
Give the user the assembled prompt in a fenced code block, ready to paste into Claude Code (/goal ...) or a Managed Agent call. Don't add commentary inside the block — commentary goes outside it.
If any of steps 1-4 couldn't be filled from the source material (goal too vague, no real constraints given, etc.), say so explicitly and ask for that piece — don't invent it.