collab

Have multiple LLMs collaboratively brainstorm solutions, then synthesize the best ideas into a plan. The LLMs build on each other's ideas across rounds rather than critiquing positions.

Load the consult-llm skill before proceeding — it defines the invocation contract (stdin heredoc, flags, output format, multi-turn). Do not call the CLI without loading it first.

Available models

Selectors resolvable in this environment (depends on configured API keys):

!`consult-llm models`

Arguments: $ARGUMENTS

Model flags: any --<selector> from the Models block above selects a collaborator (e.g. --gemini, --openai, --deepseek). Repeat for multiple. Need at least two. Translate model flags and defaults according to the loaded consult-llm skill's model-selection rules.

Strip all flags from arguments to get the task description. Use the selector name as the label when presenting per-model output.

Phase 0: Load consult-llm Skill

Load it now. Follow its invocation contract for all CLI calls in this workflow.

Phase 1: Understand the Task (No Questions)

1. Explore the codebase - use Glob, Grep, Read to understand:

   • Relevant files and their structure
   • Existing patterns and conventions
   • Dependencies and interfaces

   Before planning or consulting, do enough research to understand how the requested behavior actually works. Before starting, think about what resources would be useful to obtain first: relevant source files, tests, logs, generated files, config, examples, command output, external docs, or authoritative upstream source. Gather the cheapest useful evidence before forming a plan.

   Do not stop at the first plausible file, definition, setting, or example. Follow references, callers, related tests, and runtime usage until you can explain the current behavior and the likely impact of changing it.

2. Ground external semantics before planning - understand the requested behavior in the real system, not just this repo
   • If the task depends on an external product, CLI, API, protocol, file format, or ecosystem convention, verify the relevant behavior using the cheapest authoritative evidence available: local binaries/flags, generated files, official docs, public source, package/library code, or web search.
   • Capture only decision-relevant facts that affect scope, acceptance criteria, compatibility, or implementation constraints.
   • Do not create a separate research artifact unless the evidence materially changes the plan.
3. Make evidence-backed assumptions - do NOT ask clarifying questions
   • Use best judgment based on codebase and external context
   • Prefer simpler solutions when ambiguous
   • Follow existing patterns in the codebase
4. Prepare context summary - create a brief summary of:
   • The task to be implemented
   • Relevant files discovered
   • Key patterns and conventions in the codebase
   • Any constraints or considerations

Phase 2: Initial Ideas

Have all selected LLMs independently brainstorm approaches (in parallel).

Seed prompt:

I need to implement the following task:

[Task description]

Here's what I found in the codebase:
[Context summary - relevant files, patterns, conventions]

Brainstorm implementation ideas:
1. **Ideas**: List 2-3 possible approaches with brief descriptions
2. **Favorite**: Which approach do you lean toward and why?
3. **Open questions**: What aspects are you unsure about or would benefit from another perspective?
4. **Risks**: What could go wrong or be tricky?

Think creatively. Share rough ideas — we're exploring, not committing.

Invoke consult-llm with -f <path> for each relevant source file, sending the seed prompt per the consult-llm invocation contract. If explicit collaborator flags were supplied, pass one -m <selector> per collaborator. Otherwise omit -m so consult-llm applies configured defaults. All models are queried in parallel in a single call.

Extract per-model thread IDs from the response — needed for Phase 3 since each model receives a different prompt.

Present each set of ideas to the user, labeled by selector.

Phase 3: Build On Each Other

Each round, share every other LLM's ideas with each model and ask them to build on them (in parallel). Pass each LLM's thread ID via -t <id> to continue its conversation. Continue until the ideas converge into a clear approach — typically 2-3 rounds, but use as many as needed.

Build-on prompt (same template for each model; embed every other model's previous-round response, labeled by selector):

Your collaborator(s) shared these ideas:

[Other LLMs' responses from the previous round, each labeled with the selector name]

Build on their thinking:
1. **What resonates**: Which ideas are strong? Why?
2. **Combinations**: Can any ideas be combined into something better?
3. **New ideas**: Did their thinking spark any new approaches?
4. **Refinements**: How would you improve the most promising ideas so far?
5. **Concerns resolved**: Did their ideas address any open questions?

Keep building — don't tear down. Refine toward the best solution.

Each model receives a different prompt (the other models' responses embedded). Invoke consult-llm once with one --run flag per collaborator, continuing each model's thread.

Present every response to the user after each round, labeled by selector.

When to stop: All collaborators are refining details rather than introducing new ideas, and a clear approach has emerged. Don't stop while there are still unresolved open questions or competing directions.

Phase 4: Synthesize

After all rounds, synthesize the brainstorm into a plan:

1. Identify the strongest ideas — which approaches gained momentum across rounds?
2. Note convergence — where did the LLMs naturally align?
3. Pick the best combination — merge the strongest elements into one coherent approach
4. Write the plan:

# [Feature Name] Implementation Plan

**Goal:** [One sentence describing what this builds]

## Brainstorm Summary

**Key ideas** (one block per collaborator, labeled with the selector):
- **<selector>:** [2-3 bullet points]

**Convergence:** [Where they naturally agreed]
**Synthesis:** [How the final approach combines the best ideas]

---

### Task 1: [Short description]

**Files:**
- Create: `exact/path/to/file.py`
- Modify: `exact/path/to/existing.py` (lines 123-145)

**Steps:**
1. [Specific action]
2. [Specific action]

**Code:**
```language
// Include actual code, not placeholders
```

---

Guidelines:

• Exact file paths - never "somewhere in src/"
• Complete code - show the actual code
• Small tasks - 2-5 minutes of work each
• DRY, YAGNI - only what's needed

Save the plan to history/plan-<feature-name>.md.