Brainstorming
Use for generating or comparing multiple technical approaches to one validated decision.
Boundary: mk:office-hours answers “should we build this?”; this skill answers “how should we build it?” A bare “brainstorm” request is ambiguous: ask whether the user wants product discovery/validation or technical approaches. Do not route from that word alone.
Profiles
- quick /
--depth quick (default) — restate the decision in one line; give 2-4 technically distinct options and one recommendation. If a material fact is missing, state the assumption rather than running discovery. Return inline and stop. Do not load techniques, scout, scoring, challenge, reports, plans, wiki candidates, memory, or handoff.
- deep /
--depth deep — load the needed technique and use the deep workflow below. A report, HTML output, or handoff is optional and only produced when requested or needed by an active plan.
Routing
| Situation |
Route |
| New product, feature, or side-project value is unvalidated |
mk:office-hours |
| Observed failure or root-cause work |
mk:investigate |
| Existing plan, verdict, or brainstorm needs critique |
mk:plan-ceo-review or mk:elicit |
| Generic Q&A or a single implementation detail |
answer directly or use mk:docs-finder |
Defaults
Use at least two alternatives. Quick returns 2-4; deep generates 3-8 and never more than 8. Keep multiple dimensions in one run when they serve one decision or report; split only 3+ independently shippable concerns.
Quick Workflow
- Restate the validated decision and any material assumption.
- Give 2-4 technically distinct options with the key trade-off for each.
- Recommend one option and return inline. Do not create a report, plan, wiki candidate, or memory entry.
Deep Workflow
- Confirm the decision, binding constraint, success criterion, and excluded scope. Ask at most three focused questions when needed.
- If 3+ independently shippable concerns are bundled, ask to choose one. Use a compact scout summary only when current codebase touchpoints change the decision.
- Load one technique from
references/techniques/; generate 3-8 technical approaches without scoring while generating.
- Run one anti-bias pivot, then the challenge pass in
references/challenge-pass.md before recommending.
- Load scoring only for deep scoring. Write a report, HTML output, or handoff only when requested or needed by an active plan. For a requested report, an optional
mewkit wiki handoff suggest follows .cursor/skills/wiki/references/terminal-handoff-advisory.md; it is fail-open and never runs for quick.
--advice (composable, off by default)
Opt-in strategic supervision for one run, deep workflow only. GUIDE fires after the
frame and before generation, REVIEW after the challenge pass and before the
recommendation reaches the user; hard cap 4.
Athena supplies decision criteria and reads the chosen direction. It never generates the
option set, scores it, chooses for the user, or adds a gate.
Without the flag there are zero calls and no state is written.
Checkpoints: references/advice-checkpoints.md. Contract: .cursor/rules/domain-advice-supervision.mdc (load only when the
flag is present).
References
references/context-budget.md — deep clarification, count, pivot, and challenge limitsreferences/techniques/ — choose one technique for a deep runreferences/challenge-pass.md and references/scoring-criteria.md — deep convergence onlyreferences/editorial-html.md — HTML only for a requested reportreferences/advice-checkpoints.md — --advice checkpoint boundaries, caps, and what supervision may not touch
1---2name: mk-brainstorming-33description: Generate or compare multiple technical approaches for one validated decision. NOT for product discovery or refining an existing plan.4---56# Brainstorming78Use for generating or comparing multiple technical approaches to one validated decision.910**Boundary:** `mk:office-hours` answers “should we build this?”; this skill answers “how should we build it?” A bare “brainstorm” request is ambiguous: ask whether the user wants product discovery/validation or technical approaches. Do not route from that word alone.1112## Profiles1314- **quick** / `--depth quick` (default) — restate the decision in one line; give 2-4 technically distinct options and one recommendation. If a material fact is missing, state the assumption rather than running discovery. Return inline and stop. Do not load techniques, scout, scoring, challenge, reports, plans, wiki candidates, memory, or handoff.15- **deep** / `--depth deep` — load the needed technique and use the deep workflow below. A report, HTML output, or handoff is optional and only produced when requested or needed by an active plan.1617## Routing1819| Situation | Route |20| --- | --- |21| New product, feature, or side-project value is unvalidated | `mk:office-hours` |22| Observed failure or root-cause work | `mk:investigate` |23| Existing plan, verdict, or brainstorm needs critique | `mk:plan-ceo-review` or `mk:elicit` |24| Generic Q&A or a single implementation detail | answer directly or use `mk:docs-finder` |2526## Defaults2728Use at least two alternatives. Quick returns 2-4; deep generates 3-8 and never more than 8. Keep multiple dimensions in one run when they serve one decision or report; split only 3+ independently shippable concerns.2930## Quick Workflow31321. Restate the validated decision and any material assumption.332. Give 2-4 technically distinct options with the key trade-off for each.343. Recommend one option and return inline. Do not create a report, plan, wiki candidate, or memory entry.3536## Deep Workflow37381. Confirm the decision, binding constraint, success criterion, and excluded scope. Ask at most three focused questions when needed.392. If 3+ independently shippable concerns are bundled, ask to choose one. Use a compact scout summary only when current codebase touchpoints change the decision.403. Load one technique from `references/techniques/`; generate 3-8 technical approaches without scoring while generating.414. Run one anti-bias pivot, then the challenge pass in `references/challenge-pass.md` before recommending.425. Load scoring only for deep scoring. Write a report, HTML output, or handoff only when requested or needed by an active plan. For a requested report, an optional `mewkit wiki handoff suggest` follows `.cursor/skills/wiki/references/terminal-handoff-advisory.md`; it is fail-open and never runs for quick.4344## `--advice` (composable, off by default)4546Opt-in strategic supervision for one run, **deep workflow only**. GUIDE fires after the47frame and before generation, REVIEW after the challenge pass and before the48recommendation reaches the user; hard cap 4.4950Athena supplies decision criteria and reads the chosen direction. It never generates the51option set, scores it, chooses for the user, or adds a gate.5253Without the flag there are zero calls and no state is written.5455Checkpoints: `references/advice-checkpoints.md`. Contract: `.cursor/rules/domain-advice-supervision.mdc` (load only when the56flag is present).5758## References5960- `references/context-budget.md` — deep clarification, count, pivot, and challenge limits61- `references/techniques/` — choose one technique for a deep run62- `references/challenge-pass.md` and `references/scoring-criteria.md` — deep convergence only63- `references/editorial-html.md` — HTML only for a requested report64- `references/advice-checkpoints.md` — `--advice` checkpoint boundaries, caps, and what supervision may not touch
ngocsangyem