founder-effort

# Founder Effort A single-question calibrator that sets the **model routing tier** for the session. Every downstream skill that routes to Opus vs. Sonnet vs. an external executor reads this decision. Set it once; change it if the situation changes mid-session. > Not to be confused with Claude Code's built-in `/effort` command, which sets the current model's > *reasoning effort* (thinking depth) and never changes which model runs. This skill was renamed > from `effort` to `founder-effort` to avoid that collision. **Announce at start:** "Using founder-effort to set the token budget tier." --- ## The question Run `AskUserQuestion` with exactly one question: ``` header: "Token budget" question: "What's the effort tier for this session?" options: - label: "Ample — Opus 4.8 (Recommended for new work)" description: "Full capability throughout. Best for planning, architectural decisions, high-stakes reviews." - label: "Moderate — Sonnet 4.6" description: "Strong capability, lower cost. Right for most feature work, research, and iteration." - label: "Constrained — Sonnet + external executor" description: "Sonnet for planning and short tasks; hand heavy implementation off as a written plan file to an external executor (a cheaper Claude session, Haiku 4.5, or a third-party plan-runner) to preserve remaining tokens." - label: "Sprint end — plans only" description: "Tokens nearly exhausted. Fable writes plans only; all implementation goes to an external executor. No generation here." ``` --- ## After the answer Announce the routing decision in one line: | Tier | Announce | |---|---| | Ample | "Effort: Ample. Routing planning and implementation to **Opus 4.8**." | | Moderate | "Effort: Moderate. Planning → **Opus 4.8**, implementation → **Sonnet 4.6**." | | Constrained | "Effort: Constrained. All work → **Sonnet 4.6**; heavy lifts → **an external executor** (plan file handoff)." | | Sprint end | "Effort: Sprint end. Plans only. Implementation → **external executor** (hand it the plan file)." | Save to `~/.claude/session-context.md` (append or overwrite the `effort_tier` line). --- ## Mid-session downgrade If you observe token pressure mid-session (context approaching limits, user mentions it), call `/founder-effort` again and re-announce. Downstream skills will adjust routing automatically. --- ## Related API primitives (don't confuse them with this skill) This skill routes **which model** handles each role for the session. The Claude API has two *per-request* controls with similar names: | Primitive | What it controls | |---|---| | `output_config.effort` (`low`/`medium`/`high`/`xhigh`/`max`) | Thinking depth and token spend within one request | | Task Budgets (`output_config.task_budget`, beta `task-budgets-2026-03-13`, min 20,000 tokens) | A model-visible token countdown for a whole agentic loop — the model self-moderates | Useful, officially documented fact for the Constrained tier: on Fable 5, **lower effort settings still perform very well — often exceeding the `xhigh`/`max` performance of previous models** — so dialing effort down is a legitimate budget lever before downgrading models. --- ## Model reference | Model | Best for | Token cost | |---|---|---| | Opus 4.8 | Complex reasoning, planning, architectural review | High | | Sonnet 4.6 | Feature implementation, iteration, most tasks | Medium | | External executor (any plan-runner) | Executing a written plan file with minimal Claude token spend | Minimal | See: [Claude models overview](https://platform.claude.com/docs/en/docs/about-claude/models/overview) · [Interactive mode / AskUserQuestion](https://code.claude.com/docs/en/interactive-mode)