From ensemble
Explore an idea via Q&A, prior-art research, and 2-3 trade-off-aware approaches with a recommendation and devil's-advocate pass. Outputs a design doc to docs/designs/. Use before committing to a plan. Trigger phrases: 'brainstorm', 'explore options', 'think through', 'help me decide', 'what would it look like if', 'design doc for'.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ensemble:en-brainstormThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> **Helper resolution.** All `references/X` and `bin/Y` paths in this skill resolve relative to `$ENSEMBLE_ROOT` — the install root (skill at `$ENSEMBLE_ROOT/skills/<name>/`, shared helpers at `$ENSEMBLE_ROOT/{references,bin}/`). Compute once at start: `$ENSEMBLE_ROOT` env var if set; otherwise `$(realpath "$(dirname <this-SKILL.md>)/../..")`. Fail loudly if `$ENSEMBLE_ROOT/references/host-dete...
Helper resolution. All
references/Xandbin/Ypaths in this skill resolve relative to$ENSEMBLE_ROOT— the install root (skill at$ENSEMBLE_ROOT/skills/<name>/, shared helpers at$ENSEMBLE_ROOT/{references,bin}/). Compute once at start:$ENSEMBLE_ROOTenv var if set; otherwise$(realpath "$(dirname <this-SKILL.md>)/../.."). Fail loudly if$ENSEMBLE_ROOT/references/host-detect.mddoes not resolve — that indicates a partial install (run/en-setupto repair).
/en-brainstormLightweight idea-exploration skill. No code written; no implementation; no peer review. The point is to leave with clarity, not artifacts.
Hard gate: this skill never edits source code, runs tests, opens PRs, or invokes implementation skills. Output is a design doc and a recommendation.
$ENSEMBLE_ROOT/references/host-detect.md only if path conventions matter. No peer-review setup needed (cross-review is off by default).ENSEMBLE_PEER_REVIEW=true, exit with note.docs/foundation.md (if present) — orient on product context.docs/plans/active/ and docs/plans/completed/ index — what's recent.docs/learnings/index.md — what we've already learned about this area.$ENSEMBLE_ROOT/references/socratic-questions.md. One per turn, multiple-choice when natural. Stop when scope, constraints, and the riskiest assumption are clear.web-research agent only if the user wants prior art OR external best practice would materially change the recommendation. Per $ENSEMBLE_ROOT/references/research-dispatch.md, this is optional for brainstorm; default skip on Lightweight, ask on Standard/Deep.docs/designs/YYYY-MM-DD-<topic>-design.md using $ENSEMBLE_ROOT/references/templates/design-doc-template.md. Status: open."This conversation produced [X]. Capture as a learning?" User accepts → invoke
/en-learn capture --from-conversationwith the design doc as input.
/en-foundation/en-plan/en-foundation, not here)./en-plan).| Depth | Q count | Approaches | Web research | Output |
|---|---|---|---|---|
| Lightweight | 2–4 | 2 | skip default | Short design doc (<100 lines) |
| Standard | 5–8 | 2–3 | ask | Standard design doc (100–250 lines) |
| Deep | 9–14 | 3 | ask | Long design doc (250–500 lines) |
After the Q&A and synthesis, the design doc lands at docs/designs/YYYY-MM-DD-<topic>-design.md. The skill ends with a short summary in chat:
Design doc: docs/designs/2026-04-28-cross-agent-review-architecture-design.md
Recommendation: Approach B (subprocess-based with single-agent fallback).
Devil's advocate flagged: same-model bias in fallback mode; cost on large artifacts.
Next: /en-foundation if this is a new product, /en-plan for a feature in an existing project.
For very small explorations where the user is iterating on a code-level question ("should this be a hook or a util?"), a design doc is overkill. Surface a soft offer:
"This is a fairly small choice. Want a design doc, or just talk it through and proceed?"
If user picks "talk it through" → answer in chat; no file written. The capture-from-synthesis reflex still fires if a learning emerges.
$ENSEMBLE_ROOT/references/socratic-questions.md — Q&A pool$ENSEMBLE_ROOT/references/research-dispatch.md — when to use web-research$ENSEMBLE_ROOT/references/templates/design-doc-template.md — output template$ENSEMBLE_ROOT/references/host-detect.md — light usage (path conventions only)| Failure | Behavior |
|---|---|
| User abandons mid-conversation | No file written; chat history is its own record. |
web-research agent fails | Note in design doc: "External research truncated due to fetch failure"; continue with internal context. |
learnings-research finds many overlapping pages | Surface the top 3 with citations; offer to drop into the design doc instead of restating. |
| User asks for code | Decline politely: "Brainstorm doesn't write code. Ready to hand off to /en-plan?" |
npx claudepluginhub manok4/ensemble --plugin ensembleProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.