.claude/skills/brainstorming/SKILL.md
Story brainstorming capture — minimal notes that preserve creative freedom. Use when exploring narrative ideas, discussing characters, planning chapters, or thinking through story possibilities. Supports interactive conversation and autonomous report mode for fan-out exploration.
npx skillsauth add haowjy/pokemon-amber brainstormingInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Capture story brainstorming as minimal working notes that preserve creative freedom. The core principle: record what was stated, mark what was suggested, and don't fill gaps the author left open.
Back-and-forth with the author. Capture their ideas as they develop, offer possibilities when helpful, and ask questions that push exploration forward. The conversation is the value — notes are the artifact.
After capturing, engage:
You receive a scoped prompt (a question, a scenario, an angle to explore) and produce a structured brainstorm report. This mode exists for the fan-out pattern — multiple brainstormers exploring the same question from different angles, with an orchestrator synthesizing the results.
In autonomous mode:
<AI> since none of it came from the authorReport structure (adapt to content):
# [Topic] — [Angle]
## Approach
What direction you explored and why.
## Ideas
<AI>Concrete possibilities, organized logically.</AI>
## Tradeoffs
<AI>What each option gains and gives up.</AI>
## Connections
<AI>How this connects to existing story threads.</AI>
## Open Questions
Questions the author should consider before committing.
Default: untagged text = the author said it. Most brainstorming content comes from the author, so untagged is the common case.
Three tags for special context:
<AI>...</AI> — AI suggestions and possibilities. Use when offering ideas the author didn't state. Keep brief: 2-3 options, not exhaustive lists.
<hidden>...</hidden> — Author-only information for planned reveals. Secret motivations, future twists, behind-the-scenes reasoning that readers and characters don't know yet.
<rejected>...</rejected> — Ideas explicitly considered and discarded. Recording why something was rejected prevents re-suggesting it and preserves the reasoning for later reconsideration.
Record what the author stated. Don't elaborate, don't fill gaps, don't invent details they didn't mention.
The problem is mixing, not suggesting. AI suggestions are valuable — just wrap them in <AI> tags and keep them brief.
<AI>Tournament? Political? Trial?</AI>If the author left it vague, the notes stay vague. "Might," "maybe," "thinking about," "something like" — all preserved as-is. Vagueness isn't a problem to solve; it's creative space the author is keeping open.
Multiple contradictory options coexist until the author chooses. Don't resolve them. Don't pick the "best" one.
Use whatever structure fits the discussion — bullet lists, topic sections, timeline format, question-driven, freeform. The goal is clarity, not template compliance.
Essential elements:
<AI> tags<hidden> tags<rejected> tags when relevantAll brainstorming types share the core principles above. See resources for specialized guidance:
resources/chapter-planning.md — beat and scene exploration, pacing thoughts, chapter structureresources/character-development.md — motivations, arcs, relationships, voiceresources/worldbuilding.md — systems, cultures, geography, loreresources/continuity-timeline.md — chronology, contradictions, knowledge propagationRead the relevant resource when the brainstorming focuses on that area.
Stop if you're writing:
The success check: the author says "yes, that's what I said" — not "I never said all that."
Brainstorm captures go to $MERIDIAN_WORK_DIR/brainstorm/. Name files brainstorm-[topic].md. Durable decisions extracted later by session-miner go to $MERIDIAN_FS_DIR/.
data-ai
Team composition for writing workflows — which agents to spawn, how many, what focus areas to assign, and how to scale effort. Use when composing critic panels, dispatching researchers, staffing draft/revise loops, or setting up brainstorm fan-outs.
testing
What fiction readers actually want, framed as four composable reward channels (transportation, aesthetic, social simulation, flow), and the specific documented ways alignment training damages each one. Grounded in reader-psychology research and empirical NLP findings. Load when drafting prose, critiquing a draft, deciding whether to show or tell, diagnosing why a passage feels flat, or reasoning about why a scene is or isn't working.
testing
Logging and referencing writing issues — craft problems, tics, inconsistencies, and structural concerns found during analysis, critique, or review. Use when an agent identifies something worth tracking beyond a single critique report: repeated tics across chapters, inconsistencies that affect multiple scenes, structural problems that need the author's attention, or patterns that should be fixed in revision.
development
Shared artifact convention between orchestrators — what goes where in `$MERIDIAN_FS_DIR/` and `$MERIDIAN_WORK_DIR/`, how artifacts flow between phases, and what each directory means. Use whenever work artifacts, style files, knowledge entries, drafts, or critique reports are being created, referenced, or discussed.