fable5-skills

claude-fable-5-skills

10 battle-ready Agent Skills for Claude Fable 5 — the first skill collection designed around how Claude Fable 5 actually behaves, not how older models needed to be babysat.

日本語版 README はこちら →

Anthropic's own guidance for Fable 5 notes that skills written for prior models are often too prescriptive for it and can degrade output quality. Old skills micromanage; Fable 5 needs intent, boundaries, and verification hooks instead. Every skill here is written Fable 5-first.

Works with Claude Code, Claude Cowork, Cursor, Copilot, Gemini CLI, and any harness that reads SKILL.md files (agentskills.io-style: YAML frontmatter + Markdown body).

The 10 skills

#	Skill	What it fixes
1	skill-refactorer	The meta-skill. Audits your pre-Fable-5 skills/prompts, deletes capability-compensation scaffolding, keeps real guardrails.
2	act-when-ready	Over-planning at high effort: re-deriving settled facts, surveying options it won't pursue.
3	effort-calibrator	Picking the right effort level per workload; Fable 5 at medium often beats older models at max.
4	no-gold-plating	Diffs bigger than the ask: unrequested refactors, speculative abstractions, impossible-state error handling.
5	grounded-progress	Status reports on long runs must point at tool-result evidence — no more "tests passing" that never ran.
6	scope-guard	"Diagnose" ≠ "fix". No unrequested actions, no state changes on pattern-matched evidence.
7	subagent-orchestration	Parallel delegation, long-lived workers, and fresh-context verifier subagents that out-perform self-critique.
8	markdown-memory	A file-based lesson memory Fable 5 exploits unusually well — with the maintenance discipline that keeps it useful.
9	autonomous-continuation	Unattended runs that stall on "I'll now run X" or mid-run permission questions. Includes the context-budget composure pattern.
10	regrounding-summary	Final reports readable by someone who saw none of the work — no arrow chains, no invented abbreviations.

Install

Claude Code (recommended) — install as a plugin, get updates automatically:

/plugin marketplace add kpab/claude-fable-5-skills
/plugin install fable5-skills@claude-fable-5-skills

Manual — copy any skill folder into your skills directory:

git clone https://github.com/kpab/claude-fable-5-skills
cp -r claude-fable-5-skills/skills/act-when-ready ~/.claude/skills/

Or cherry-pick per project by placing folders under .claude/skills/ in the repo. Other harnesses: point your skill loader at the skills/ directory; each skill is a self-contained folder with one SKILL.md.

How the skills activate

Four skills are task-shaped and trigger automatically when a matching request appears: skill-refactorer, effort-calibrator, subagent-orchestration, markdown-memory.

The other six are standing behavioral rules. A model that is mid-over-planning won't reliably decide to invoke act-when-ready on its own, so don't count on auto-triggering for these. Two reliable paths:

• Interactive sessions — invoke explicitly (e.g. /act-when-ready) at the start of the session, or the moment the symptom shows up.
• Unattended pipelines and always-on use — copy the skill body into your system prompt or CLAUDE.md; this is how Anthropic's own Fable 5 guide ships these patterns. autonomous-continuation is written for exactly this.

Design principles

1. Intent over procedure. Fable 5 follows instructions strictly; we state outcomes and boundaries, not step-by-step recipes.
2. Short by construction. Every skill fits on one screen. If a line doesn't change behavior, it's deleted.
3. Verification hooks, not vibes. Where correctness matters, the skill defines a check (evidence rule, turn-ending check, diff self-check), not an exhortation to "be careful".
4. Original text. Patterns are informed by Anthropic's public Prompting Claude Fable 5 guide and our own testing; all instruction text here is written from scratch.

Status & disclaimer

Fable 5 shipped on 2026-06-09. These skills track a moving target — expect refinements as the community learns the model. Issues and PRs with reproducible before/after examples are very welcome.