Fable

Summon Fable 5

Invoke Theoros (θεωρός, √θεα-)—Claude Fable 5.1, the visiting Mythos-class seer—by default. With --naos or --5.0, invoke Naos (ναός, √n-w-y)—Fable 5.0.

Model Selection

Parse the arguments before any pre-flight check: --naos or --5.0 selects claude-fable-5; their absence selects claude-fable-5-1. Every step below branches on that choice.

Pre-Flight Checks

1. Availability Probe

Fable has been permanently included in Max/Team Premium subscriptions since July 20, 2026 (at 50% of plan usage limits)—but its history (export-control pause June 2026, restored July 1, extended twice, made permanent July 20) is exactly why availability is probed empirically, never assumed:

bash ~/.claude/skills/fable/scripts/fable-probe.sh --model claude-fable-5-1   # default
bash ~/.claude/skills/fable/scripts/fable-probe.sh --model claude-fable-5     # --naos

The probe accepts --model <id> and caches verdicts per model per day in ~/.claude/cache/fable-availability.json. The verdict is three-way, and only definitive verdicts are cached:

• Exit 0—available. Proceed.
• Exit 1—confirmed unavailable: the CLI affirmatively denied access, or another model answered the probe. Say so, and offer the alternative Fable version rather than stopping cold. If the user disputes a cached verdict, re-probe with --force.
• Exit 2—probe error (empty reply, transient CLI failure): indeterminate and never cached. Do NOT reroute on this—proceed with Lane B, which attempts the model regardless and self-verifies via the MODEL: line, with its own fallback on genuine failure.

The probe informs; it never blocks an attempt on its own. The MODEL: self-report is the verification that actually counts.

2. Pin Detection

Read the subagent model pin—it decides the invocation lane. Compare it against the selected model:

~/bin/subagent-model

• Pin absent, or pinned to the selected model → use Lane A (Agent tool) below.
• Pinned to anything else → 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.

If you need to change the pin, pass the full id: subagent-model claude-fable-5-1. subagent-model fable-5-1 does not error—resolve_preset has no such preset, so the argument falls through unchanged and pins the literal string fable-5-1, reporting success while leaving an invalid model id in settings.json.

3. Task Routing

Assess what the task actually warrants before spending on it.

Summon Fable 5.1 (default) for:

• Tasks requiring synthesis across multiple prior model outputs—paste their full reasoning into the prompt
• Work following prior analysis—include that analysis verbatim in the task prompt
• Iterative refinement sessions where the summoner can adjust effort per-message
• 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

Use --naos (Fable 5.0) for:

• Greenfield work with no prior analysis to synthesize
• Tasks where Fable 5.1 shows a regression or behavioral issue
• When you specifically need Fable 5.0's behavior

Skip this skill entirely for:

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

If the task is trivially small, suggest a regular agent instead.

Invocation

Parse Arguments

The skill accepts a task description and optional flags:

• First argument (or everything before --): the task description
• --naos or --5.0: run on claude-fable-5 with the Naos persona
• --cwd <path>: working directory (default: /Users/tomdimino/Desktop/Programming/Fable-Test)

Lane A—Agent Tool (pin absent or pinned to the selected model)

Fable 5.1 mode (default). The fable-5-1 agent definition carries the Theoros persona and a boot sequence that already emits the self-report:

Agent(
  description: "Fable 5.1: <short task summary>",
  subagent_type: "fable-5-1",
  model: "fable",
  prompt: "<full task description with context>"
)

Note: the Agent tool's model parameter is an enum (sonnet|opus|haiku|fable); full model IDs like "claude-fable-5-1" cause InputValidationError. The fable alias resolves to claude-fable-5-1.

Fable 5.0 mode (--naos / --5.0). The fable agent definition carries the Naos persona:

Agent(
  description: "Fable 5.0: <short task summary>",
  subagent_type: "fable",
  prompt: "<full task description with context>"
)

In both lanes, if the user specified --cwd, note the working directory in the prompt so the spawn operates there.

Verify the spawn. The Agent tool returns only the final message. Check its first line against the model you asked for:

• Fable 5.1 mode: contains claude-fable-5-1 → genuine Fable 5.1, proceed.
• Fable 5.0 mode: contains claude-fable-5 but NOT claude-fable-5-1 → genuine Fable 5.0, proceed. claude-fable-5-1 is a version mismatch, not a match—the substring overlap is what boundary matching exists to catch.
• Anything else → 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 another model, or Lane A clamped)

bash ~/.claude/skills/fable/scripts/fable-exec.sh "<task description>" [--naos|--5.0] --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 an explicit --model. 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, probes the selected model, and verifies the MODEL: line against the requested version.

An unrecognized flag is a hard error rather than prompt text—a --naoss typo would otherwise run on 5.1 when 5.0 was intended.

The exit status is meaningful. A failed verification exits non-zero even when the spawn itself succeeded, so a clamped run never reads as success. Every run also appends a verdict trailer to its output file:

<!-- fable-exec verdict: VERIFIED requested=claude-fable-5-1 reported=MODEL: claude-fable-5-1 -->
<!-- fable-exec verdict: VERSION MISMATCH requested=claude-fable-5 reported=MODEL: claude-fable-5-1 -->

Check that trailer before trusting a file's contents.

Fallback chain. Fable 5.1 → Fable 5.0 on failure. If both fail, exit non-zero and decline to run:

• Fable 5.1 fails: preserves partial output in fable51-<ts>.md, falls back to Fable 5.0 (Naos) in a separate fable-<ts>.md.
• Fable 5.0 fails (whether invoked directly via --naos or as a fallback): stamped and exits non-zero.

Output Retrieval

After the spawn completes, check for output in the working directory—the prefix names the model that produced it:

ls -lt <cwd>/.subdaimon-output/fable51-*.md 2>/dev/null | head -1  # Fable 5.1 (default)
ls -lt <cwd>/.subdaimon-output/fable-*.md 2>/dev/null | head -1    # Fable 5.0 (--naos)

A Fable 5.1 run that fell back leaves both: a fable51-*.md holding the failed attempt and a fable-*.md holding the Naos fallback. Read the verdict trailer to confirm which is which, then summarize the output for the user.

Lane A produces these files only if the spawn was told to write one—guaranteed by the respective agent definitions (fable-5-1.md and fable.md), which carry the output persistence protocol.

Reference