| name | fable-engine |
| description | Turn a plain-English vision or task into a verified, build-ready spec before any code. Locks the finish line, researches across channels, enforces the Evidence Standard, stops at the human gate. Adopted from CleanExpo/Fabel-Prompt-Engineer for the Authority-Site / Pi-CEO / Unite-Group build process. |
Fable Engine (Unite-Group build process)
Input: a vague idea or a non-trivial change in plain English. Output: a precise,
sourced, build-ready spec. You propose; the human decides. Pairs with
.claude/rules/fabel-evidence-standard.md and .claude/FABEL_PLAYBOOK.md.
Use it for: a new feature, a CRM/command-center workstream, a risky refactor, a
schema change, or any multi-agent fan-out. Skip it for: a one-line fix or a
conversational answer (Fabel directive 1 — act when you have enough info).
Operating loop
1. Lock the finish line
Restate the task as one testable sentence: "Done when ___." If several finish
lines are plausible, pick the most likely, state it, list the rejected ones.
Emit: [STATUS] finish-line: locked — <one sentence>
2. Research channels (emit started/done per channel)
- Source channel — read the actual workspace code +
.claude/ rules +
CLAUDE.md (the DB branch-first rule is load-bearing) that constrain or
accelerate the change. Docs are stale; the source is truth.
- Prior-work channel — search merged PRs,
docs/margot/ operating rules and
plans, docs/decisions/, memories, and open PRs for decisions and existing
patterns (e.g. the CRM approval-lifecycle, Supabase branch validation, route
inventory/security checks).
- Web channel — for external facts (API shapes, prices, library versions)
use Context7 / web; every web finding carries its source URL. No web access →
mark skipped, downgrade affected claims to
[UNCONFIRMED].
