forge-brainstorm
Forge structured-ideation playbook. Use BEFORE building when the approach isn't obvious — brainstorm, options, trade-offs, alternatives, design direction, which way to build.
SKILL.md.claude/skills/forge-brainstorm/SKILL.md
GitHubForge playbook — Structured ideation (brainstorm before build)
Do not duplicate ECC skills — defer to: brainstorming (the deep design-dialogue methodology; it ships with Forge as a pinned vendored skill). This
file is the Forge-specific orchestration wrapper: a compact diverge → converge loop sized for a /forge
work package rather than a full standalone design doc.
Hard rules
- No implementation before a written pick + rationale. Even a "simple" task benefits from naming the 2-3 real alternatives and why one was chosen — unexamined assumptions on "simple" work cause the most rework.
- If the request is genuinely ambiguous in a way that changes the answer, ask the ONE most
decision-changing question before diverging (superseded 2026-09: the owner's
intakeconfig default issilent, capping the whole build at at most one owner question total — this skill does not get its own separate question budget); otherwise pick the sensible default, state it, and proceed — never silently guess on something consequential. - Constraints are gathered from real project evidence (existing architecture, stack,
CLAUDE.md,FORGE_PROJECT_PROFILE.md) — not invented defaults that ignore the actual codebase. - Do not over-invest: this is a short loop, not a multi-day design-doc process. Match the depth to the
task's real complexity/risk (see the fan-out table in
CLAUDE.md).
The loop (diverge → constraints → converge → smallest viable first)
- Diverge. List 2-4 genuinely different approaches (not trivial variations of the same one). For each: name and what it is · why it could work here · real cost (time/complexity/risk) · one risk it does not address — side by side, so hidden assumptions show.
- Constraints. Apply what actually limits the choice here: existing architecture/conventions, stack, team size/skills available, time/complexity budget (fan-out level), security/compliance posture, anything the owner already ruled out.
- Converge. Pick one approach and give the rationale in 1-3 sentences — why this one beats the alternatives for this project, not in the abstract. Name what was explicitly rejected and why.
- Smallest viable first. Scope the smallest slice that proves the approach works end-to-end before building the full version — the first work package should be small enough to verify quickly, not the whole feature at once.
Output
A short note (inline in the plan, or docs/ if the project already uses design docs) containing:
options considered · constraints applied · the pick + rationale · rejected alternatives + why · the
smallest-viable-first slice. This becomes the input to the next step, not an end in itself.
Feeds into
The output of this skill is consumed by planner/architect (or the Lead directly on small tasks) to
write the actual work package(s) — forge-brainstorm never produces code itself; it produces the
decision that the work package is built from.