Diverge — Divergent Option-Generation Engine
Break premature convergence. Autoregressive reasoning commits to the first plausible answer; this skill manufactures genuine breadth before any evaluation, then converges deliberately.
Core principle: Separate generation from judgment. Ideas are generated in ISOLATION (no branch sees another, so none can anchor on another), and only a SEPARATE critic pass is allowed to evaluate. Evaluating while generating is the failure mode this skill exists to prevent.
Pre-Flight Gate
Activation is decided by the frontmatter description alone — this file is read only AFTER the skill
has been selected. Nothing here can prevent over-triggering, so this gate is a post-selection
redirect, not a filter.
Before proceeding, check the ABORT conditions below. If any holds, STOP, say in one line which one
fired, and take the redirect instead of fanning out any agents. Being invoked — including a literal
/diverge, "brainstorm options", "explore alternatives", or "give me alternatives" — does NOT
override an ABORT: the phrasing that selected this skill is not evidence the problem is open.
| ABORT when the ask is… |
Redirect |
| Syntax, a definition, a factual lookup, arithmetic, a deterministic transform |
Answer directly |
| A bug with a single known root cause |
/investigate |
| A decision the user has already made |
/execute — just build it |
| Following an explicit step-by-step instruction |
Do the steps |
| Low-stakes or adjacent — naming, wording, two known libraries |
Answer directly with the inline obvious/alternative/trap (anti-anchoring rule) |
| Consequential but CLOSED — one approach is clearly right |
State it, plus a one-line trap check |
| Open but cheaply reversible — trying one costs less than exploring |
Pick the obvious one and try it |
PROCEED only when no ABORT fired AND consequence × openness both hold — hard to reverse, wide
blast radius, or high cost of being wrong, AND several viable, structurally different approaches
plausibly exist. The gate is consequence × openness, NOT token cost.
Phase 1: DIVERGE — Isolated Parallel Generation
- Restate the problem in one sentence and list the hard constraints (these go to every frame).
- Pick ~5 frames from the table below whose vantage best surfaces structurally different approaches for THIS problem. Use 4 for narrow problems, up to 6–7 for wide ones. More than 7 hits diminishing returns (see
multi-agent-orchestration).
- Send a single message with multiple
Agent tool calls for parallel agent launches — one Agent per frame, all in the same message. Each runs in isolation and never sees the others.
- Give every sub-agent the SAME generator-only instruction:
You are a GENERATOR working under the "" frame. From this vantage only:
PROBLEM: CONSTRAINTS:
Produce 3–5 genuinely distinct approaches. For each: a one-line name, the core idea (2–3 sentences), and the single biggest risk.
RULES: Do NOT evaluate, rank, hedge, or pick a winner. Do NOT mention other frames or approaches (you cannot see them). Favor range over safety — at least one approach must be non-obvious. Output a flat list. No preamble.
Isolation invariant: No generator receives another's output. No generator is asked to compare, score, or choose. Violating this re-introduces the anchoring the skill exists to remove.
Gate: All frame agents returned. You hold a flat, unscored, un-deduplicated pool. No evaluation has happened yet.
Phase 2: FOCUS — Separate Critic Pass
Only now is judgment allowed. Do this yourself (or one critic agent) over the WHOLE pool.
- Score each approach 1–5 on three axes, then weighted total:
score = 0.25·novelty + 0.45·viability + 0.30·fit
(novelty = distance from the obvious default; viability = buildable under constraints; fit = serves the real goal)
- Flag traps — high-novelty/seductive approaches that fail a hard constraint. List each with the SPECIFIC constraint it violates. Traps are reported, never silently dropped.
- Cluster survivors by underlying angle (collapse near-duplicates from different frames; note where isolated frames converged — that convergence is signal).
- Deepen the top 3 by weighted score: mechanism, key trade-off, first failure it would hit, rough cost/effort.
Gate: Every approach scored; traps flagged with reasons; clusters formed; top 3 deepened.
Output Shape
- Brief — one line: the problem and the constraints that bind it.
- Wide set — full pool as a flat list, each with a chip
[N x.x | V x.x | F x.x → total].
- Converge — shortlist — top 3, with the single best non-obvious pick STARRED (★) and one line on why it beats the obvious default. List traps separately below, each with the constraint it violates.
- Deepened branches — the mechanism/trade-off/first-failure/cost for the top 3.
- One provocation — a single closing line: the assumption most worth challenging, or the next option if constraints loosened. Exactly one.
Anti-Patterns
| Bad Habit |
What /diverge Does Instead |
| Commit to the first plausible answer |
Forces a wide isolated pool before any judgment |
| Generators that rank or hedge themselves |
Generator-only instruction; judgment is a separate pass |
| Branches seeing each other's ideas |
Isolation invariant — parallel Task calls, no shared context |
| Running on closed/low-stakes problems |
Pre-flight gate aborts unless consequence AND openness hold |
| Dropping the seductive-but-wrong option silently |
Traps flagged with the exact constraint violated |
| Ending with a menu of equals |
One starred non-obvious pick + one provocation |
| 12 frames "to be thorough" |
~5 frames; diminishing returns past 7 |
Integration with Other Skills
/diverge is the divergent complement to the convergent /deep-analysis.
| Situation |
Use |
| Open-ended, consequential — "what are our options?" |
/diverge (this skill) |
| One approach chosen — pressure-test it |
/deep-analysis |
| Bug with unknown root cause |
/investigate |
| Winners chosen — build them |
/execute |
| Cheap inline judgment call |
anti-anchoring rule habit — no skill needed |
Hand-off chain: /diverge (widen) → pick winners → /deep-analysis (validate) → /execute (build). Never feed the raw wide set to /execute; converge first.
Cost Note
The Diverge phase spends ~5 parallel sub-agents plus one critic pass — real but bounded. Justified ONLY by the pre-flight gate (consequence × openness), never by default. On a closed or low-stakes problem this skill is WRONG, not merely expensive: it manufactures noise and false choices.
Frames Table
| Frame |
Vantage prompt |
| Regulator |
"What does compliance, audit, security, least-privilege demand? Design for the reviewer and the breach." |
| Attacker |
"How would I break, abuse, or exploit this? Design the approach that survives an adversary." |
| Inversion |
"What guarantees failure? Now invert each failure into a design constraint." |
| First-principles |
"Ignore prior art. From the raw requirement and cost/physics, what is the minimal thing that works?" |
| Markets / incentives |
"Who pays, who benefits, what behavior does this reward? Follow the incentives and the cost curve." |
| Biology / evolution |
"Treat it as an organism: how does it adapt, self-heal, degrade gracefully, scale without central control?" |
| Operator / 3am |
"I am on-call when this breaks. What is observable, reversible, boring to run? Optimize for the worst night." |
| Scale-10x |
"Requirements hold but load/data/team are 10x. Which approaches survive, which shatter?" |
| Constraint-removal |
"Which 'hard' constraint is actually assumed? Remove the most load-bearing one — what opens up?" |
| Future-maintainer |
"Someone inherits this in two years with no context. What is simplest to understand, change, and delete?" |
References
anti-anchoring rule — the always-on inline habit this skill is the heavy mode fordeep-analysis skill — the convergent partner; validates the chosen winnermulti-agent-orchestration skill — parallel-launch idiom; 5–7-agent diminishing-returns limitexecute skill — builds the converged winners
1---2name: diverge3description: Structured divergence engine for consequential, open-ended decisions. Generates a wide, deliberately non-anchored option space by spawning isolated parallel sub-agents under different cognitive frames, then runs a separate critic pass that scores, declusters, and deepens the best. The divergent complement to /deep-analysis. Warranted only when a decision is both hard to reverse or wide in blast radius AND genuinely open with several structurally different answers — architecture and platform strategy, design-space exploration, de-anchoring a premature conclusion, or a high-stakes what are our options. Triggers include diverge, brainstorm options, explore alternatives, option space, divergent thinking. Because it fans out many agents in parallel it is not for cheap or adjacent asks — naming, wording, picking between two known libraries, or any question with one defensible answer should be answered directly.4---56# Diverge — Divergent Option-Generation Engine78Break premature convergence. Autoregressive reasoning commits to the first plausible answer; this skill manufactures genuine breadth before any evaluation, then converges deliberately.910**Core principle:** Separate generation from judgment. Ideas are generated in ISOLATION (no branch sees another, so none can anchor on another), and only a SEPARATE critic pass is allowed to evaluate. Evaluating while generating is the failure mode this skill exists to prevent.1112## Pre-Flight Gate1314Activation is decided by the frontmatter description alone — this file is read only AFTER the skill15has been selected. Nothing here can prevent over-triggering, so this gate is a **post-selection16redirect**, not a filter.1718**Before proceeding, check the ABORT conditions below. If any holds, STOP, say in one line which one19fired, and take the redirect instead of fanning out any agents.** Being invoked — including a literal20`/diverge`, "brainstorm options", "explore alternatives", or "give me alternatives" — does NOT21override an ABORT: the phrasing that selected this skill is not evidence the problem is open.2223| ABORT when the ask is… | Redirect |24|---|---|25| Syntax, a definition, a factual lookup, arithmetic, a deterministic transform | Answer directly |26| A bug with a single known root cause | `/investigate` |27| A decision the user has already made | `/execute` — just build it |28| Following an explicit step-by-step instruction | Do the steps |29| Low-stakes or adjacent — naming, wording, two known libraries | Answer directly with the inline obvious/alternative/trap (`anti-anchoring` rule) |30| Consequential but CLOSED — one approach is clearly right | State it, plus a one-line trap check |31| Open but cheaply reversible — trying one costs less than exploring | Pick the obvious one and try it |3233**PROCEED only when no ABORT fired AND consequence × openness both hold** — hard to reverse, wide34blast radius, or high cost of being wrong, AND several viable, structurally different approaches35plausibly exist. The gate is consequence × openness, NOT token cost.3637## Phase 1: DIVERGE — Isolated Parallel Generation38391. Restate the problem in one sentence and list the hard constraints (these go to every frame).402. Pick ~5 frames from the table below whose vantage best surfaces structurally different approaches for THIS problem. Use 4 for narrow problems, up to 6–7 for wide ones. More than 7 hits diminishing returns (see `multi-agent-orchestration`).413. **Send a single message with multiple `Agent` tool calls for parallel agent launches** — one Agent per frame, all in the same message. Each runs in isolation and never sees the others.424. Give every sub-agent the SAME generator-only instruction:4344> You are a GENERATOR working under the "<FRAME>" frame. From this vantage only:45> PROBLEM: <one-sentence problem> CONSTRAINTS: <hard constraints>46> Produce 3–5 genuinely distinct approaches. For each: a one-line name, the core idea (2–3 sentences), and the single biggest risk.47> RULES: Do NOT evaluate, rank, hedge, or pick a winner. Do NOT mention other frames or approaches (you cannot see them). Favor range over safety — at least one approach must be non-obvious. Output a flat list. No preamble.4849**Isolation invariant:** No generator receives another's output. No generator is asked to compare, score, or choose. Violating this re-introduces the anchoring the skill exists to remove.5051**Gate:** All frame agents returned. You hold a flat, unscored, un-deduplicated pool. No evaluation has happened yet.5253## Phase 2: FOCUS — Separate Critic Pass5455Only now is judgment allowed. Do this yourself (or one critic agent) over the WHOLE pool.56571. **Score** each approach 1–5 on three axes, then weighted total:58 `score = 0.25·novelty + 0.45·viability + 0.30·fit`59 (novelty = distance from the obvious default; viability = buildable under constraints; fit = serves the real goal)602. **Flag traps** — high-novelty/seductive approaches that fail a hard constraint. List each with the SPECIFIC constraint it violates. Traps are reported, never silently dropped.613. **Cluster** survivors by underlying angle (collapse near-duplicates from different frames; note where isolated frames converged — that convergence is signal).624. **Deepen the top 3** by weighted score: mechanism, key trade-off, first failure it would hit, rough cost/effort.6364**Gate:** Every approach scored; traps flagged with reasons; clusters formed; top 3 deepened.6566## Output Shape67681. **Brief** — one line: the problem and the constraints that bind it.692. **Wide set** — full pool as a flat list, each with a chip `[N x.x | V x.x | F x.x → total]`.703. **Converge — shortlist** — top 3, with the single best non-obvious pick STARRED (★) and one line on why it beats the obvious default. List **traps separately** below, each with the constraint it violates.714. **Deepened branches** — the mechanism/trade-off/first-failure/cost for the top 3.725. **One provocation** — a single closing line: the assumption most worth challenging, or the next option if constraints loosened. Exactly one.7374## Anti-Patterns7576| Bad Habit | What `/diverge` Does Instead |77|-----------|------------------------------|78| Commit to the first plausible answer | Forces a wide isolated pool before any judgment |79| Generators that rank or hedge themselves | Generator-only instruction; judgment is a separate pass |80| Branches seeing each other's ideas | Isolation invariant — parallel Task calls, no shared context |81| Running on closed/low-stakes problems | Pre-flight gate aborts unless consequence AND openness hold |82| Dropping the seductive-but-wrong option silently | Traps flagged with the exact constraint violated |83| Ending with a menu of equals | One starred non-obvious pick + one provocation |84| 12 frames "to be thorough" | ~5 frames; diminishing returns past 7 |8586## Integration with Other Skills8788`/diverge` is the **divergent** complement to the **convergent** `/deep-analysis`.8990| Situation | Use |91|-----------|-----|92| Open-ended, consequential — "what are our options?" | `/diverge` (this skill) |93| One approach chosen — pressure-test it | `/deep-analysis` |94| Bug with unknown root cause | `/investigate` |95| Winners chosen — build them | `/execute` |96| Cheap inline judgment call | `anti-anchoring` rule habit — no skill needed |9798**Hand-off chain:** `/diverge` (widen) → pick winners → `/deep-analysis` (validate) → `/execute` (build). Never feed the raw wide set to `/execute`; converge first.99100## Cost Note101102The Diverge phase spends ~5 parallel sub-agents plus one critic pass — real but bounded. Justified ONLY by the pre-flight gate (consequence × openness), never by default. On a closed or low-stakes problem this skill is WRONG, not merely expensive: it manufactures noise and false choices.103104## Frames Table105106| Frame | Vantage prompt |107|-------|----------------|108| **Regulator** | "What does compliance, audit, security, least-privilege demand? Design for the reviewer and the breach." |109| **Attacker** | "How would I break, abuse, or exploit this? Design the approach that survives an adversary." |110| **Inversion** | "What guarantees failure? Now invert each failure into a design constraint." |111| **First-principles** | "Ignore prior art. From the raw requirement and cost/physics, what is the minimal thing that works?" |112| **Markets / incentives** | "Who pays, who benefits, what behavior does this reward? Follow the incentives and the cost curve." |113| **Biology / evolution** | "Treat it as an organism: how does it adapt, self-heal, degrade gracefully, scale without central control?" |114| **Operator / 3am** | "I am on-call when this breaks. What is observable, reversible, boring to run? Optimize for the worst night." |115| **Scale-10x** | "Requirements hold but load/data/team are 10x. Which approaches survive, which shatter?" |116| **Constraint-removal** | "Which 'hard' constraint is actually assumed? Remove the most load-bearing one — what opens up?" |117| **Future-maintainer** | "Someone inherits this in two years with no context. What is simplest to understand, change, and delete?" |118119## References120121- `anti-anchoring` rule — the always-on inline habit this skill is the heavy mode for122- `deep-analysis` skill — the convergent partner; validates the chosen winner123- `multi-agent-orchestration` skill — parallel-launch idiom; 5–7-agent diminishing-returns limit124- `execute` skill — builds the converged winners
pfangueiro