sumitaich1998/jarvisvr / skills/orchestration/result-synthesis
result-synthesis
VerifiedMerge the specialists' results into one coherent spoken reply and coordinate the final spatial layout, closing out a multi-agent turn. Use after the team's sub-tasks finish (or partially fail) to produce a single agent.speech{final} and hand compositing to the stage-agent. Triggers: synthesize, summarize the results, wrap up, combine answers, final reply, "put it all together".
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.2 KB |
SKILL.md
View on GitHub---
name: result-synthesis
description: >-
Merge the specialists' results into one coherent spoken reply and coordinate
the final spatial layout, closing out a multi-agent turn. Use after the team's
sub-tasks finish (or partially fail) to produce a single agent.speech{final}
and hand compositing to the stage-agent. Triggers: synthesize, summarize the
results, wrap up, combine answers, final reply, "put it all together".
license: MIT
compatibility: >-
Requires the JarvisVR agent-backend orchestrator; coordinates stage-agent for
holo.* layout (protocol v1.2).
metadata:
agent: jarvis
category: orchestration
version: "1.0"
author: jarvisvr
allowed-tools: arrange_holograms
---
# Result Synthesis
Close the loop on an `orchestration.plan`: gather each agent's result, resolve
conflicts, speak **one** answer, and make sure the holograms they spawned are
arranged comfortably (not piled on top of each other).
## Inputs
- The `plan_id` and the per-agent results (each agent's `data` + the `holo.spawn`
objects it produced, addressed by `object_id`).
- The final `orchestration.agent_status{state:"done"|"failed"}` for every node.
## Steps
1. **Collect & order results** in the order most useful to the user (usually:
what they asked first → supporting detail). Use each tool's `data.speech`
suggestion as raw material, not verbatim.
2. **Reconcile conflicts / dedupe.** If two agents answered overlapping
questions, keep the higher-confidence / more specific one.
3. **Compose one spoken reply.** Stream long answers as multiple
`agent.speech{final:false}` and end with a single `agent.speech{final:true}`.
Keep it conversational; don't read JSON aloud.
4. **Coordinate the layout.** Prefer delegating to the `stage-agent`
(`compose-workspace`) so it decides anchoring/arrangement. For a quick gather
you may emit `holo.layout` yourself via `arrange_holograms`.
5. **Mark the turn done** — emit `agent.thinking{stage:"done"}` after the final
speech.
## Output examples
Spoken synthesis (after perception + research + productivity finished):
```json
{ "text": "That's your coffee mug. Tokyo is 18° and cloudy, and your 5-minute timer is running.",
"final": true, "emotion": "neutral" }
```
Final layout via the stage compositor (`arrange_holograms` → `holo.layout`,
protocol §5.10):
```json
{ "arrangement": "arc", "anchor": "head", "spacing": 0.25,
"objects": ["O_vision_annotation", "O_weather_orb", "O_timer"] }
```
## Partial failure
If a node reports `state:"failed"`, synthesize around it honestly — report what
succeeded and name what didn't, without dropping the whole turn:
```json
{ "text": "Tokyo is 18° and cloudy and your timer's set. I couldn't get a clear
look at the desk object though — want me to try again?", "final": true }
```
## Edge cases
- **Nothing succeeded** → one apologetic `agent.speech{final:true}` + an optional
`notify` toast; do not spawn data widgets.
- **Too many holograms** → ask `stage-agent` to run `declutter-space` before the
final arrangement.
- **Barge-in** (`client.barge_in`, §5.14) arrived mid-synthesis → stop streaming
speech, keep already-spawned holograms, end the turn.
- **Streaming order** → never emit `final:true` until every kept result is voiced.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |