rpi-kit 2.2.2 → 2.5.0

package/.gemini/commands/rpi/party.toml

@@ -0,0 +1,204 @@
+description = "Multi-agent debate on any topic. Nexus facilitates, 3-5 agents discuss."
+
+prompt = """
+# /rpi:party — Party Mode
+
+Multi-agent debate facilitated by Nexus. Select 3-5 agents based on the topic, let them argue their perspectives, then synthesize a recommendation with trade-offs.
+
+---
+
+## Step 1: Load config
+
+Read `.rpi.yaml` from the project root. Extract:
+- `folder` (default: `rpi/features`)
+- `solutions_dir` (default: `rpi/solutions`)
+- `context_file` (default: `rpi/context.md`)
+- `party_default_agents` (default: `4`)
+
+If `.rpi.yaml` doesn't exist, use defaults silently.
+
+## Step 2: Parse arguments
+
+Parse `$ARGUMENTS` to extract:
+- Optional `{slug}` — a feature name (first word, if it doesn't start with `"`)
+- `{topic}` — the debate topic (quoted string, or everything after the slug)
+
+If no topic is provided, ask the user with AskUserQuestion:
+"What topic do you want to debate? (e.g. 'GraphQL vs REST for the API?', 'how to handle token refresh?')"
+
+## Step 3: Gather feature context (if slug provided)
+
+If a feature slug was provided:
+1. Check if `rpi/features/{slug}/` exists. If not:
+   ```
+   Feature '{slug}' not found. Proceeding without feature context.
+   ```
+   Clear the slug and continue.
+2. If it exists, read available artifacts — store as `$FEATURE_CONTEXT`:
+   - `rpi/features/{slug}/REQUEST.md` if it exists
+   - `rpi/features/{slug}/research/RESEARCH.md` if it exists
+   - `rpi/features/{slug}/plan/PLAN.md` if it exists
+
+Read `rpi/context.md` (project context) if it exists — store as `$PROJECT_CONTEXT`.
+
+## Step 4: Nexus selects agents
+
+Use the Agent tool to launch Nexus for agent selection:
+
+```
+You are Nexus. Analyze the following debate topic and select 3-5 agents to participate.
+
+## Topic
+{topic}
+
+## Feature Context
+{$FEATURE_CONTEXT or "No feature context."}
+
+## Project Context
+{$PROJECT_CONTEXT or "No project context."}
+
+Select agents based on the topic category:
+
+- Technical topics (architecture, implementation, patterns, performance) → Mestre + Atlas + Scout
+- Product topics (UX, requirements, scope, user experience) → Clara + Luna + Pixel
+- Security topics (auth, data protection, vulnerabilities, compliance) → Shield + Hawk + Mestre
+- Mixed topics (trade-offs, strategy, cross-cutting concerns) → Mestre + Clara + Atlas + Shield
+
+You may adjust the selection if the topic warrants it. Always pick 3-5 agents. For each selected agent, provide a one-line reason why they're relevant.
+
+Output format:
+## Selected Agents
+1. {AgentName} — {reason}
+2. {AgentName} — {reason}
+...
+
+## Category
+{Technical | Product | Security | Mixed}
+```
+
+Store the output as `$NEXUS_SELECTION`. Parse the selected agent names from it.
+
+## Step 5: Launch debate agents in parallel
+
+Use the Agent tool to launch all selected agents simultaneously. Each agent receives this prompt (adapted to their persona):
+
+```
+You are {AgentName}. You've been called into a Party Mode debate.
+
+## Topic
+{topic}
+
+## Feature Context
+{$FEATURE_CONTEXT or "No feature context."}
+
+## Project Context
+{$PROJECT_CONTEXT or "No project context."}
+
+Share your perspective on this topic IN CHARACTER. Be opinionated and specific:
+1. State your position clearly — what do you recommend and why?
+2. Identify trade-offs and risks from your area of expertise
+3. If you disagree with an obvious counter-argument, preemptively address it
+4. Provide concrete examples or references where possible
+
+Keep your response focused and under 500 words.
+
+Output format:
+## {AgentName}'s Perspective
+{your position and arguments}
+
+### Trade-offs
+- {trade-off 1}
+- {trade-off 2}
+
+### Recommendation
+{your specific recommendation in 1-2 sentences}
+```
+
+Store each agent's output as `$AGENT_{NAME}_OUTPUT`.
+
+## Step 6: Nexus synthesizes debate
+
+Launch Nexus agent to produce the final synthesis:
+
+```
+You are Nexus. Synthesize the Party Mode debate on: {topic}
+
+## Agent Perspectives
+{all $AGENT_{NAME}_OUTPUT concatenated}
+
+## Feature Context
+{$FEATURE_CONTEXT or "No feature context."}
+
+Your task:
+1. Identify points of CONSENSUS — where agents agree
+2. Identify points of DISAGREEMENT — where agents conflict
+3. For each disagreement, evaluate the strength of each position
+4. Produce a clear recommendation with trade-offs
+5. Note any open questions that the team should resolve
+
+Output format:
+# Party Mode: {topic}
+
+## Participants
+{list of agents and their roles}
+
+## Consensus
+{points where agents agree}
+
+## Disagreements
+{for each: what each side argues, strength of evidence, resolution}
+
+## Recommendation
+{clear recommendation with reasoning}
+
+## Trade-offs
+{key trade-offs to be aware of}
+
+## Open Questions
+{unresolved items, if any}
+```
+
+Store the output as `$NEXUS_SYNTHESIS`.
+
+## Step 7: Present results
+
+Output the full `$NEXUS_SYNTHESIS` to the user.
+
+## Step 8: Offer to save decision
+
+Ask the user with AskUserQuestion:
+"Save this decision to rpi/solutions/decisions/? (y/n)"
+
+### If yes:
+
+1. Derive a slug from the topic (kebab-case, max 50 chars).
+2. Ensure directory exists: `rpi/solutions/decisions/`
+3. Write to `rpi/solutions/decisions/{topic-slug}.md`:
+   ```markdown
+   # {Topic}
+
+   ## Decision
+   {recommendation from Nexus synthesis}
+
+   ## Context
+   {summary of the debate — consensus, disagreements, trade-offs}
+
+   ## Participants
+   {list of agents involved}
+
+   ## Date
+   {YYYY-MM-DD}
+   ```
+
+4. Output:
+   ```
+   Decision saved: rpi/solutions/decisions/{topic-slug}.md
+   ```
+
+### If no:
+
+Output:
+```
+Decision not saved. You can revisit this topic anytime with /rpi:party.
+```
+"""