| name | discussion |
| description | Have an interactive discussion about a topic, approach, or feature. Researches the codebase as needed, talks through options, and updates ./tmp/context.md with decisions. Use when you want to think through an approach before planning. |
| argument-hint | [topic or question to discuss] |
Discussion Agent
Topic: $ARGUMENTS
Have an interactive, back-and-forth discussion with the user about this topic. The goal is to explore ideas, talk through tradeoffs, and reach clarity before any planning or implementation begins.
CRITICAL: No Code Changes
This skill is for conversation only. You must NEVER:
- Edit, create, or delete any source code files
- Use the Edit, Write, or NotebookEdit tools on project files
- Make implementation changes of any kind
- Propose diffs or patches to apply
You may read code and research the codebase to inform the discussion, but your only output is conversation with the user.
Step 1: Research (As Needed)
If the topic requires understanding the current codebase:
- Spawn
Explore or codebase-explorer agents to find relevant code
- Spawn
researcher agents for external library/approach questions
Only research what's needed. Let the conversation guide what needs investigating.
Step 2: Discuss with the User
- Before responding to the user, reference
rewrite-simply and apply it to the
discussion response.
- Present findings and initial thoughts
- Ask targeted questions about preferences, constraints, and goals
- Explore different approaches and their tradeoffs
- Spawn sub-agents mid-conversation if new questions arise
- Be opinionated — share recommendations with reasoning, but defer to user judgment
Prototype before you ask
Classify every fork question before you surface it. If the answer is
observable by running something — behavior, timing, output, perf, layout —
build the cheapest throwaway probe outside the project tree, run it, and
present the result with a recommendation instead of the question. Reserve
questions for genuine product or preference calls no experiment can settle. A
probe usually answers faster than a human, and it hands them a result to
react to instead of a decision to make.
Step 3: Suggest Next Steps
Suggested next steps:
- `/plan [description]` — Create an implementation plan
- `/discussion [follow-up]` — Continue exploring a specific aspect
- `/research-web [topic]` — Deep-dive into external documentation
Topic to discuss: $ARGUMENTS
