| name | diverge |
| description | Generate 3-5 alternative framings of a problem BEFORE committing to one. Use upstream of `quality-planner` when the task admits multiple credible approaches and choosing the wrong shape would cost rework. Distinct from prometheus (interview to clarify WHAT to build), metis (stress-test a draft plan), oracle (debug second opinion), and abstraction-critic (critique an existing artifact's paradigm fit). Divergent-framer expands the option space; the others narrow it. |
| allowed-tools | Agent |
/diverge — Expand the option space before commit
Use /diverge <task> when the task admits multiple credible framings and you want to see them side-by-side with explicit tradeoffs before a planner commits to one.
When to use
- Open-ended prompts: "how should we approach the rate-limit handling?", "what's the best architecture for this?", "is there a better way to do X?"
- Architecturally significant choices where the paradigm decision is non-obvious (event-sourced vs CRUD, push vs pull, sync vs async, monolith vs services, library-build vs use-existing).
- The model has stated an interpretation and you want to verify it covers the space before commit.
- A council surfaced findings that disagree on framing — the lenses are pointing at different shapes of solution.
When NOT to use
- Single-line bug fixes, well-scoped refactors, or any task with an obvious dominant approach. Adding
/diverge to routine work is friction, not value.
- Tasks already mid-execution.
/diverge is upstream-only — once a planner has committed to a paradigm, use metis to stress-test it or abstraction-critic to question the fit.
- Debugging. Use
oracle for "what's the root cause?" and metis for "what's wrong with this plan?".
How it works
The skill dispatches the divergent-framer agent with your task. The agent emits a structured menu:
- 3-5 framings, each with: a 2-4 word name, the mental model in one sentence, what the paradigm makes EASY (1-2 concrete affordances), what it makes HARD (1-2 concrete costs), and the best-fit conditions.
- A one-paragraph rank picking the strongest fit for the stated constraints with a "redirect if" clause naming the condition under which a different framing would win.
The agent does NOT produce implementation plans (that's quality-planner / prometheus) or critique existing code (that's quality-reviewer / metis). The output is the menu, not the recipe.
Decision tree — divergent vs convergent critics
| Question | Skill | What it produces |
|---|
| What shapes COULD this problem take? | /diverge (divergent-framer) | 3-5 framings + rank |
| WHAT exactly does the user want me to build? | /prometheus | Interview-clarified requirements |
| Is THIS plan robust against edge cases? | /metis | Plan stress-test findings |
| What's the root cause / best path through this stuck debug? | /oracle | Diagnostic + approach recommendation |
| Does this artifact's paradigm match the problem? | /abstraction-critic | Paradigm-fit critique on existing code |
/diverge runs upstream of all of them. The others narrow the answer; /diverge expands the option space.
Example flow
/diverge handling rate-limit retries in the resume-watchdog
→ divergent-framer emits:
Framing 1: pinned-cooldown-with-jitter — easy: deterministic; hard: thundering herd
Framing 2: exponential-backoff-with-cap — easy: handles transient; hard: harder to reason about
Framing 3: queue-then-replay — easy: durable; hard: requires storage
Framing 4: skip-and-notify-only — easy: simple; hard: user must act
Rank: Framing 1 because watchdog cooldown is already pinned; Framing 3 if durability becomes a hard requirement.
You pick Framing 1. quality-planner gets a clean target instead of starting from "I think we should...".
Why this exists
oh-my-claude's existing critics (oracle, metis, abstraction-critic, excellence-reviewer) are convergent — they narrow toward a single right answer. That convergence is correct after the framing is chosen. Before the framing is chosen, the model defaults to the first paradigm that comes to mind — usually the most-recently-seen example or the easiest to articulate, not necessarily the best fit.
/diverge is the seam that protects against premature commitment. It's the most-explicit application of the user's "natural divergent thinking ability" axis-6 ask.