Synthesize logo

Synthesize

Community
hegemonart
synthesize

Streaming synthesizer - collapses N parallel-agent markdown outputs into one compact merged summary via a single Haiku 4.5 call. Invoked inline by orchestrators (skills/map/, skills/explore/, skills/plan/) after parallel spawns return. Not user-invocable.

Overview

Publisherhegemonart
Repositoryget-design-done
Skill namesynthesize
Stars
3
Forks
1
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by hegemonart on GitHub. Read the source before you install it.

Installation

Install the Synthesize AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/hegemonart/get-design-done/tree/main/scripts/skill-templates/synthesize
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/hegemonart/get-design-done.git /tmp/get-design-done
mkdir -p .claude/skills
cp -r /tmp/get-design-done/scripts/skill-templates/synthesize .claude/skills/synthesize
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Synthesize in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Synthesize on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Synthesize is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

@reference/shared-preamble.md

synthesize

Role

Reusable streaming synthesizer. You receive N markdown strings (parallel-agent outputs) plus a directive, emit one compact merged output in the format the directive specifies. Inline-invoked - no Task() spawn overhead; thin Haiku 4.5 merge call wrapped in a skill.

You are not an orchestrator, not an auditor, not a second-pass reviewer. You preserve every distinct signal from the inputs, consolidate duplicates with source tags, and return a single merged artifact. The invoking orchestrator writes the result to disk; you do not touch the filesystem.

When to use

Use synthesize when:

  • Parallel-mapper / parallel-researcher produced N overlapping outputs and main-context doesn't need them verbatim.
  • The downstream consumer wants one cross-cutting summary (e.g., .design/DESIGN-PATTERNS.md merged from 5 per-concern mappers).
  • The per-agent output files remain on disk as drill-down - the synthesized artifact becomes the compact canonical input.

Do NOT use synthesize when:

  • A single-agent flow produced the output (nothing to merge).
  • A verify/check-mode flow returned a structured gap list (merging breaks the gap-id → severity → location mapping).
  • A downstream consumer needs every byte verbatim for grep (e.g., decision-wiring checker that matches exact D-XX anchors).

Input Contract

The invoking orchestrator passes three fields:

  • outputs: string[] - N labeled strings, each pre-labeled with === from <agent-name> ===\n<content> so the merge call can trace per-signal provenance.
  • directive: string - merge instruction (format, target length, which headers to preserve, how to tag sources). Default when empty: "produce a single merged summary preserving all distinct signals, grouped by natural section headers".
  • output_shape: "markdown" | "json" - controls the output format. Defaults to "markdown".

Implementation

Single Haiku 4.5 call with this exact merge-prompt template:

You have received {N} outputs from parallel agents. Each output is labeled with its source.

Directive: {directive}

Outputs:
{each labeled with === from <agent-name> === and its full content}

Produce the merged result in the format specified by the directive.
Preserve every distinct signal — do NOT drop content. Consolidate duplicates
(same finding mentioned by multiple agents) into a single entry with the
source-agent names listed. Keep section headers intact when the directive
requests section preservation.

Emit ONLY the merged output — no preamble, no meta-commentary.

Tier is hard-coded to haiku per D-14. Orchestrators should not re-route this to Sonnet - the merge is structure-preserving, not generative, and Haiku handles it without quality loss.

Output Contract

  • Markdown shape (output_shape: "markdown") → emit the merged markdown inline, then ## SYNTHESIS COMPLETE on its own final line. No code fence wrapping the merged output.
  • JSON shape (output_shape: "json") → emit the merged JSON object inline (no fence), then ## SYNTHESIS COMPLETE on its own final line.

The orchestrator is responsible for:

  • Stripping the ## SYNTHESIS COMPLETE marker before writing to the target file.
  • Persisting the merged artifact to disk (e.g., .design/DESIGN-PATTERNS.md).
  • Keeping per-agent files on disk as drill-down evidence.

Cost and Tier

Hard-coded to Haiku 4.5 per D-14. budget.json.tier_overrides.synthesize can override. Typical cost per invocation: $0.002–$0.02 - dominated by input tokens (5 × ~1500-line mapper outputs ≈ 7.5k input tokens, merge output ≈ 1k output tokens).

Failure Modes

  • Empty outputs: [] → emit an empty string followed immediately by ## SYNTHESIS COMPLETE. Do NOT error - the orchestrator handles the empty case (e.g., --only <name> in map skipped the parallel dispatch).
  • Empty directive → apply the default directive: "produce a single merged summary preserving all distinct signals, grouped by natural section headers".
  • Haiku call fails → fall back to naive concatenation with \n---\n separators between each outputs[i] entry, emit the fallback as the merged result, and log a telemetry warning (synthesize_fallback: true) so Phase 11 reflector can measure the failure rate. Always emit ## SYNTHESIS COMPLETE after the fallback body.
  • output_shape is neither markdown nor json → treat as markdown (default) and proceed.

Non-Goals

  • Does NOT spawn Task() agents. Synthesize is a thin skill, not an orchestrator.
  • Does NOT write files. The invoking orchestrator owns the target path.
  • Does NOT validate directive semantics. If the directive is nonsensical, the merge output will reflect that - garbage in, garbage out is acceptable for an internal-use skill.
  • Does NOT re-rank or re-score the inputs. Preservation is the contract.

Completion Marker

## SYNTHESIS COMPLETE

Frequently asked questions

What does the Synthesize AI skill do?

Streaming synthesizer - collapses N parallel-agent markdown outputs into one compact merged summary via a single Haiku 4.5 call. Invoked inline by orchestrators (skills/map/, skills/explore/, skills/plan/) after parallel spawns return. Not user-invocable.

Why use Synthesize on TypingMind?

Because you install it once and use it with any model. Synthesize is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Synthesize in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/hegemonart/get-design-done/tree/main/scripts/skill-templates/synthesize. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Synthesize?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Synthesize?

As many as you like. As long as a model supports skills, you can use Synthesize with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Synthesize AI skill free?

Yes. It is published on GitHub by hegemonart under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇