fable

Summon Fable 5

Invoke Naos (ναός, √n-w-y)—Claude Fable 5, the visiting Mythos-class spirit.

Pre-Flight Checks

1. Availability Probe

Anthropic's Fable window has shifted twice (paused June 2026, restored July 1, extended through July 19)—never trust a calendar date. Probe empirically:

bash ~/.claude/skills/fable/scripts/fable-probe.sh

Exit 0 means Fable is available (verdict cached per-day in ~/.claude/cache/fable-availability.json). Exit 1 means the xenos daimon has departed—inform the user and stop. If the user disputes a cached verdict, re-probe with --force.

2. Pin Detection

Read the subagent model pin—it decides the invocation lane:

~/bin/subagent-model

• Pin absent, or pinned to claude-fable-5 → use Lane A (Agent tool) below.
• Pinned to anything else (e.g. claude-opus-4-6) → use Lane B (CLI) directly. Do NOT attempt Lane A: CLAUDE_CODE_SUBAGENT_MODEL silently overrides the Agent tool's model: parameter (anthropics/claude-code#57718)—the spawn would run on the pinned model with no signal in the tool result.

3. Task Routing

Before summoning Fable, assess whether the task warrants Mythos-class resources ($10/M input, $50/M output—2x Opus pricing).

Summon Fable for:

• End-to-end code migrations or framework upgrades
• Complex multi-file refactors (10+ files, interconnected logic)
• Architectural redesigns requiring full-system context
• Deep research synthesis with sustained reasoning
• Full-stack feature implementation from spec to tests
• Linear A decipherment pipeline work

Use Opus or Sonnet instead for:

• Quick edits, typo fixes, single-line changes
• Simple file lookups or grep searches
• Routine code review
• Anything under 5 minutes of work

If the task is clearly Opus-appropriate, suggest using a regular agent instead. If unclear, proceed—Fable handles straightforward tasks reliably too.

Invocation

Parse Arguments

The skill accepts a task description and optional flags:

• First argument (or everything before --): the task description
• --cwd <path>: working directory (default: /Users/tomdimino/Desktop/Programming/Fable-Test)

Lane A—Agent Tool (pin absent or pinned to fable)

Use the Agent tool with the model override:

Agent(
  description: "Fable: <short task summary>",
  subagent_type: "fable",
  model: "fable",
  prompt: "<full task description with context>"
)

Pass the task description as the prompt. If the user specified --cwd, note the working directory in the prompt so Fable operates there.

Verify the spawn. The fable agent's boot sequence makes it state its model id as the first line of output. Check it:

• First line contains claude-fable-5 → genuine Fable, proceed.
• Anything else (e.g. claude-opus-4-6) → the harness clamped the model silently. State the clamp explicitly to the user, then re-run the task via Lane B.

Lane B—CLI (pin set to a non-fable model, or Lane A clamped)

bash ~/.claude/skills/fable/scripts/fable-exec.sh "<task description>" --cwd "<working directory>" [--turns <N>]

--turns caps the spawn's tool calls (default 100—matching the agent definition's budget).

This spawns a fresh Claude Code process with --model claude-fable-5. The env var only governs subagent model resolution—it cannot touch a spawned process's main model, so this lane works from any parent without a restart. The script prepends its own self-report instruction and records a MODEL: line in the output file.

Output Retrieval

After Fable completes, check for output in the working directory:

ls -lt <cwd>/.subdaimon-output/fable-*.md 2>/dev/null | head -1

Read and summarize the output file for the user.

Reference