jjilli

jjilli/fable-flow

Plugin Claude Code

Marketplace for the fable-flow plugin (this repository is both the marketplace and the plugin).

jjilli/fable-flow

Plugin Claude Code

Multi-agent shipping pipeline tuned for Claude Fable 5: parallel Sonnet 5 scouts, an xhigh-effort Fable 5 architect, Fable 5 implementers in isolated git worktrees, merge, an Opus 4.8 adversarial review loop, a post-implementation bug-iteration flow, per-repo pipeline memory, a card-based direction-brainstorm before…

jjilli/fable-flow

Agent

Planning agent for the fable-flow pipeline. Delegate to the architect when a task description and scout digests are ready to be turned into an implementation plan with file-disjoint parallel tracks. The architect plans; it never edits files.

jjilli/fable-flow

Agent

Implementation agent for the fable-flow pipeline. Delegate one implementer per plan track; each runs in its own isolated git worktree and can safely execute in parallel with other implementers. Also used to apply review fixes on an integration branch.

jjilli/fable-flow

Agent

Adversarial review agent for the fable-flow pipeline. Delegate reviewers (several in parallel, each with a different lens) after tracks are merged into an integration branch, to verify the diff against the plan and requirements before a PR is opened.

jjilli/fable-flow

Agent

Read-only reconnaissance agent for the fable-flow pipeline. Delegate to scout when a fable-flow command needs codebase intelligence gathered before planning — architecture, conventions, or blast-radius for a specific task. Spawn several scouts in parallel, each with a different lens. Scouts never modify files.

jjilli/fable-flow

Command

Run only the fable-flow Explore phase — three parallel Sonnet 5 scouts (structure, conventions, blast-radius) producing digests under .fable-flow/ for a later plan.

jjilli/fable-flow

Command

Run the fable-flow Implement + Merge phases — one Fable 5 implementer per plan track, each in its own git worktree in parallel, then merge onto an integration branch and run integration verification.

jjilli/fable-flow

Command

Post-implementation iteration loop for bugs found by hand — reproduce the bug, root-cause it, fix it on the current branch with a regression test, run one correctness review pass, and bank the lesson to pipeline memory.

jjilli/fable-flow

Command

Run only the fable-flow Plan phase — a high-effort Fable 5 architect turns the scout digests into a file-disjoint parallel track plan saved to .fable-flow/plan.md.

jjilli/fable-flow

Command

Run only the fable-flow Review phase — parallel high-effort Opus 4.8 reviewers (correctness, fidelity, integration) verify a branch's diff against the plan; findings are reported, and fixed only with --fix.

jjilli/fable-flow

Command

Run the full fable-flow pipeline on a feature or fix — parallel scouts → Fable 5 plan → parallel Fable 5 implementers in worktrees → merge → Opus 4.8 review loop → branch/PR.

jjilli/fable-flow

Skill Claude CodeCodex

Battle-tested engineering patterns and seam traps distilled from long multi-round fable-flow builds. Consult when planning tracks, implementing a track, or reviewing — especially for the recurring integration failure (tests green per track, broken where tracks or the live runtime meet), the reusable patterns that make…

jjilli/fable-flow

Skill Claude CodeCodex

Reference for prompting Claude Fable 5 — how Fable prompting differs from Opus/Sonnet-era prompting, effort-level selection, and the verbatim steering snippets. Use when writing or editing prompts, subagent definitions, or slash commands that target Fable 5, or when a Fable-driven pipeline misbehaves (overplanning…

jjilli/fable-flow

Skill Claude CodeCodex

How to make frontends that look genuinely designed instead of generic "AI slop", and how to verify UI work the pipeline can't catch with a test suite. Load whenever a task builds or restyles a user-facing surface — a page, a component, a full redesign, a dashboard, a marketing site. Covers distinctive typography…

jjilli/fable-flow

Skill Claude CodeCodex

Relentless, one-question-at-a-time interview that stress-tests a plan or design before any code is written. Invoke ONLY when the user explicitly asks for it — they say "grill me", "grill the plan", "grilling", or otherwise ask to be grilled / to pressure-test the design. Do NOT auto-invoke it for ordinary planning or…