A judgment layer for Claude Code

Spend judgment first.

Token efficiency. A 30-second calibration at session start — effort tier, domain, what done looks like — governs how millions of downstream tokens get spent. Wrong-direction work and wrong-model routing are the two most expensive failure modes of an agentic session, and both are cheapest to prevent before the first tool call. The judgment layer spends a few hundred tokens to make that decision explicit.

Fable 5 is a leap in intuition. A model that can genuinely judge when to plan, what to route where, and how much effort a task deserves makes a calibration layer worth building. These skills give that intuition a structured place to act — explicit, inspectable, and reusable — instead of leaving it implicit and unrepeatable inside each session.

Three concepts.

1. Skills are the judgment layer

Claude Code ships powerful execution primitives. A skill encodes when and how to use them — under what conditions, in what order, with what routing and checks.

2. The planner is not the coder

These skills follow one orchestration pattern: the planning model writes an in-depth plan, an implementation model executes it, the planner reviews the result against the plan. The separation keeps the planner honest and the implementer focused. The /founder-effort skill sets which tier each role routes to.

3. Token budget shapes the whole session

Opus on everything is powerful and expensive; Sonnet is fast and cost-effective; a written plan file can be handed to an external executor for minimal spend. The skills make that routing decision explicit at session start — not buried inside each task.

Two commands.

/plugin marketplace add craigm26/founder-skills

# the calibrators — start here
/plugin install session-start@founder-skills
/plugin install founder-effort@founder-skills

# or all fifteen, one loop
for s in session-start founder-effort market-validation build-options prd tasks \
  fable-orchestrated-feature-dev fable-repo-audit fable-org-audit \
  fable-loop-design tufte-viz ecosystem-planning \
  skill-style-guide skill-release-gate skill-freshness-watch; do
  /plugin install ${s}@founder-skills
done

Plugins and marketplaces are a native Claude Code feature — see Plugins and Plugin marketplaces.

What /session-start looks like.

One AskUserQuestion call, three questions, ~30 seconds:

Then the routing announcement:

Effort: Focused. Planning → Opus 4.8, implementation → Sonnet 4.6.
This session I'll route to /market-validation → /build-options.
Say the skill name to begin.

Each output is the next input.

A cited evidence pack → a weighted decision matrix → an implementation-ready PRD → a machine-verifiable task plan → code, reviewed against the plan. Enter wherever your work starts: an existing feature begins at /prd; research-only ends at the evidence pack.

Walkthroughs.

Session layer

Fable 5 orchestration

Workflow chain

Craft

Maintainer tooling

AskUserQuestion.

The calibration skills are built on one tool: typed questions with described options collect better answers, faster, than freeform prompts. Wire it into your own skills the same way:

AskUserQuestion({
  questions: [{
    header: "Short label",
    question: "The actual question?",
    multiSelect: false,
    options: [
      { label: "Option A", description: "What it means" },
      { label: "Option B", description: "What it means" }
    ]
  }]
})

See Interactive mode in the official docs.

Part of a larger catalog.

Go deeper.

GitHub repository · claude-skills catalog · Claude Code documentation · Skills · Plugin marketplaces