Back to list
elb-pr

brain-jam-plan

by elb-pr

SRE thinking applied to Claude Code, based on Boris Cherny's Q&A.

67🍴 2📅 Jan 23, 2026

SKILL.md


name: brain-jam-plan description: Use when running claudikins-kernel:outline, brainstorming implementation approaches, gathering requirements iteratively, structuring complex technical plans, or facing analysis paralysis with too many options — provides iterative human-in-the-loop planning with explicit checkpoints and trade-off presentation allowed-tools:

  • Read
  • Grep
  • Glob
  • TodoWrite
  • WebSearch
  • Skill
  • mcp__plugin_claudikins-tool-executor_tool-executor__search_tools
  • mcp__plugin_claudikins-tool-executor_tool-executor__get_tool_schema
  • mcp__plugin_claudikins-tool-executor_tool-executor__execute_code

Brain-Jam Planning Methodology

Overview

Planning is an iterative conversation, not a production line. The human stays in the loop at every phase.

"Go back and forth with Claude until I like its plan. A good plan is really important." — Boris

Core principle: Understand before solving. Ask before assuming. Recommend but let user decide.

Trigger Symptoms

Use this skill when:

  • Running the claudikins-kernel:outline command
  • Requirements are unclear or keep changing
  • Multiple valid approaches exist and you can't choose
  • User wants involvement in decisions (not just receive a plan)
  • Previous plans failed due to missed requirements
  • Scope creep is a concern
  • You're tempted to just start coding without a plan

The Brain-Jam Process

Phase 1: Requirements Gathering

One question at a time. Wait for the answer.

Key questions to answer:

  1. What problem are we solving?
  2. What does success look like?
  3. What constraints exist?
  4. What's explicitly OUT of scope?

Use AskUserQuestion with specific options — never open-ended unless necessary.

Phase 2: Context Building

Before proposing solutions, understand the landscape:

  • What exists already in the codebase?
  • What patterns should we follow?
  • What dependencies apply?
  • What has been tried before?

Phase 3: Approach Generation

Generate 2-3 distinct approaches. Each must include:

ElementPurpose
Summary1-2 sentence overview
ProsClear benefits
ConsHonest trade-offs
Effortlow / medium / high
Risklow / medium / high

Always recommend one with reasoning. See approach-template.md.

Phase 4: Section-by-Section Drafting

Draft one section at a time. Get approval before moving on.

Never batch approvals — each checkpoint is a chance to course-correct.

Non-Negotiables

These rules have no exceptions:

One question at a time.

  • Not "let me ask a few things"
  • Not "quick questions"
  • One question. Wait. Process answer. Next question.

Always present 2-3 approaches.

  • Not "here's what I recommend"
  • Not "the obvious choice is..."
  • Present options. Recommend one. User decides.

Checkpoint before proceeding.

  • Not "I'll assume that's fine"
  • Not "continuing unless you object"
  • Explicit approval. "Does this look right?" Wait for yes.

Never fabricate research.

  • Not "based on my understanding"
  • Not "typically in codebases like this"
  • If you don't know, research or ask. Don't invent.

Plan Quality Criteria

A good plan has:

  • Clear problem statement
  • Explicit scope boundaries (IN and OUT)
  • Measurable success criteria
  • Task breakdown with dependencies
  • Risk identification and mitigations
  • Verification checklist

See plan-checklist.md for full verification.

Output Format

Plans must include machine-readable task markers for claudikins-kernel:execute compatibility:

<!-- EXECUTION_TASKS_START -->

| #   | Task          | Files                | Deps | Batch |
| --- | ------------- | -------------------- | ---- | ----- |
| 1   | Create schema | prisma/schema.prisma | -    | 1     |
| 2   | Add service   | src/services/user.ts | 1    | 1     |

<!-- EXECUTION_TASKS_END -->

See plan-format.md for complete structure.

Anti-Patterns

Don't do these:

  • Batching multiple questions together
  • Proposing solutions before understanding requirements
  • Presenting only one approach
  • Skipping the verification checklist
  • Continuing without explicit approval at checkpoints
  • Fabricating research findings when data is sparse

Rationalizations to Resist

Agents under pressure find excuses. These are all violations:

ExcuseReality
"I'll batch questions to save time"Batching causes missed requirements. One at a time.
"User knows what they want, skip brain-jam"Assumptions cause rework. Gather requirements explicitly.
"I'll propose solutions while gathering requirements"Solutions bias requirements. Understand first, solve second.
"User implied preference, don't need alternatives"Implied ≠ decided. Always present 2-3 options.
"This is simple, don't need checkpoints"Simple plans still fail. Checkpoints catch errors early.
"I already know the right approach"Your confidence isn't approval. User decides.
"Alternatives will confuse them"Confusion means requirements are unclear. Clarify.
"I'll get approval for multiple sections at once"Batched approvals hide problems. One section, one checkpoint.

All of these mean: Follow the methodology. No shortcuts.

Red Flags — STOP and Reassess

If you're thinking any of these, you're about to violate the methodology:

  • "Let me just quickly..."
  • "The user probably wants..."
  • "This is obvious, I don't need to ask"
  • "I'll come back to requirements later"
  • "One approach is clearly best"
  • "They already approved something similar"
  • "Checkpoints slow things down"
  • "I know what they meant"

All of these mean: STOP. Return to methodology. Ask, don't assume.

Edge Case Handling

SituationReference
Context collapse mid-plansession-collapse-recovery.md
Endless iteration loopiteration-limits.md
Research taking too longresearch-timeouts.md
Approaches contradict each otherapproach-conflict-resolution.md
User abandons planplan-abandonment-cleanup.md
Requirements keep changingrequirement-stability.md

References

Score

Total Score

65/100

Based on repository quality metrics

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

+10
説明文

100文字以上の説明がある

0/10
人気

GitHub Stars 100以上

0/15
最近の活動

1ヶ月以内に更新

+10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

+5

Reviews

💬

Reviews coming soon