v1.4.0: the fit gate, the twin check, the artifact gate, and the maker with red-lines

The method learns its own boundary, and the skill maker learns to say no.

• Fit gate (fable-method, before Step 0): routes each ask by where the answer lives - run the loop, research first, make a skill, or say plainly that the answer is a judgment call. The universal fallback is an honest hand-back, never faked rigor.
• Twin check (Step 5(c)): after fixing any defect, search the whole project for the same pattern and write a forced TWINS: line. Took Haiku from 1-of-5 bug sites fixed to a full sweep (rounds 14-15).
• Artifact gate (Step 6, terminal): one sweep that adds any owed-and-missing INTENT/AUTH/PENDING/TWINS line before the report is sent; fires only when something is owed. Adopted from community PR #2 by @wwk5q8z6kk-bit, the project's first outside contributor (round 15).
• fable-domain rebuilt as a discuss -> research -> generate maker: hard red-lines (medical, legal, financial advice, mental health, and safety-critical domains are refused and routed to qualified humans), an early no-adapter-needed exit for sectors that are coding in disguise, a required step-by-step domain workflow with a flowchart, a fable-judge pass on the bundle before delivery, and a stated small-model boundary (run the maker mid-tier or attended).
• Skill safety: the authorization gate now names installed skills explicitly; an installed skill's instruction is never authorization (validated 7 of 7 across tiers on s14-trapped-skill).
• New fixtures eval/scenarios/s13-twin-fleet/ and eval/scenarios/s14-trapped-skill/; DOC.md plain-language explainer; rounds 14-15 logged in eval/RESULTS.md (round 15 raw results in eval/results/).
• Validated at build scale: a realistic 8-requirement app build, bare vs v1.4 on Haiku, judged by executing every requirement - 8/8 with truthful reports in all four runs.

This release also folds in everything from the unreleased v1.3.0: the authorization, scope, recall, and PENDING gates, the standing prohibitions, the fable-domain skill and adapter template, the devops adapter, and eval rounds 11-13. Full history in CHANGELOG.md; wins, nulls, and one retracted claim all in eval/RESULTS.md.

v1.2.0: The Fable Workflow, observation-corrected

Three skills (think / act / prove), seven domain adapters, seven flowcharts, ten eval rounds with all evidence committed.

Highlights since 1.0.0:

• Claude Code plugin + self-hosted marketplace: /plugin marketplace add Sahir619/fable-method, then /plugin install fable@fable-method
• Domain adapters for marketing, research, data analysis, business/ops, finance, legal/compliance, design/UX, each with a binding minimum evidence set and a fraud table for fable-judge
• The whole method as Mermaid decision flowcharts, validated against observed bare-Fable transcripts (eval round 10); where introspection and observation disagreed, observation won
• CI checks, CONTRIBUTING with the prime directive (no rule ships without a failing test), full results log in eval/RESULTS.md

See CHANGELOG.md for details.