Agent skill

brain-jam-plan

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

Stars 115
Forks 6

Install this agent skill to your Project

npx add-skill https://github.com/elb-pr/claudikins-kernel/tree/main/skills/brain-jam-plan

SKILL.md

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:

Element Purpose
Summary 1-2 sentence overview
Pros Clear benefits
Cons Honest trade-offs
Effort low / medium / high
Risk low / 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:

markdown
<!-- 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:

Excuse Reality
"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

Situation Reference
Context collapse mid-plan session-collapse-recovery.md
Endless iteration loop iteration-limits.md
Research taking too long research-timeouts.md
Approaches contradict each other approach-conflict-resolution.md
User abandons plan plan-abandonment-cleanup.md
Requirements keep changing requirement-stability.md

References

  • plan-checklist.md — Verification checklist
  • approach-template.md — How to present options
  • plan-format.md — Output structure for claudikins-kernel:execute
  • iteration-limits.md — When to stop iterating
  • requirement-stability.md — Scope creep detection
  • approach-conflict-resolution.md — Conflicting approaches
  • session-collapse-recovery.md — Context collapse handling
  • research-timeouts.md — Timeout handling
  • plan-abandonment-cleanup.md — Cleanup procedures

Expand your agent's capabilities with these related and highly-rated skills.

elb-pr/claudikins-kernel

strict-enforcement

Use when running claudikins-kernel:verify, checking implementation quality, deciding pass/fail verdicts, or enforcing cross-command gates — requires actual evidence of code working, not just passing tests

115 6
Explore
elb-pr/claudikins-kernel

shipping-methodology

Use when running claudikins-kernel:ship, preparing PRs, writing changelogs, deciding merge strategy, or handling CI failures — enforces GRFP-style iterative approval, code integrity validation, and human-gated merges

115 6
Explore
elb-pr/claudikins-kernel

git-workflow

Use when running claudikins-kernel:execute, decomposing plans into tasks, setting up two-stage review, deciding batch sizes, or handling stuck agents — enforces isolation, verification, and human checkpoints; prevents runaway parallelization and context death

115 6
Explore
davila7/claude-code-templates

verl-rl-training

Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.

23,776 2,298
Explore
davila7/claude-code-templates

openrlhf-training

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

23,776 2,298
Explore
davila7/claude-code-templates

gguf-quantization

GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.

23,776 2,298
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results