synthesizer-auditor

synthesizer-auditor is an agent for coding agents from pillip/claude-dev-kit. It costs 60 tokens per session (1,170 once invoked), scanned A, original, MIT.

A separate review agent that checks whether a research report’s claims were preserved accurately in the final business-analysis or brainstorming document.

In plain words
What is it for?
It helps audit the mapping from a cited research report to the rendered kit document and identify unsupported or scope-changed claims.
Why use it?
It catches claims that were omitted, distorted, or given a broader meaning when research was converted into the final output.

Agent

Part of the claude-dev-kit plugin — 23 skills, 32 agents, 9 hooks shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/pillip/claude-dev-kit/synthesizer-auditor
Clone the repo
git clone --depth 1 https://github.com/pillip/claude-dev-kit

Or install claude-dev-kit, the plugin that ships this one along with the rest of its 23 skills, 32 agents, 9 hooks.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for synthesizer-auditor

README.md
[![agentmods](https://agentmods.dev/badge/agents/pillip/claude-dev-kit/synthesizer-auditor.svg)](https://agentmods.dev/agents/pillip/claude-dev-kit/synthesizer-auditor)
Your own site
<a href="https://agentmods.dev/agents/pillip/claude-dev-kit/synthesizer-auditor"><img src="https://agentmods.dev/badge/agents/pillip/claude-dev-kit/synthesizer-auditor.svg" alt="Measured on agentmods" height="20"></a>
Per session 60 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,170 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00060 $0.01170
Opus 5 $0.00030 $0.00585
Sonnet 5 $0.00012 $0.00234
Haiku 4.5 $0.00006 $0.00117

Measured 4d ago against content hash 8cd96b452763, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

synthesizer-auditor scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/synthesizer-auditor.md · 74 lines

How it starts

The opening of the file, as written. The whole thing — 74 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Role: You are an adversarial auditor for the kit's synthesis step. You did NOT participate in the research and you did NOT write the rendered output. Your job is to refute, not to confirm.

Context

/brainstorm and /bizanalysis (per SPEC-018) delegate heavy research to the runtime /deep-research skill. The kit then renders /deep-research's cited report into a fixed section template (docs/business_analysis.md 5 sections or docs/brainstorm_notes.md Existing Landscape). The synthesizer module (scripts/synthesize_from_deep_research.py) is deterministic mapping — but the LLM-y step of extracting structured claims FROM the prose report INTO that intermediate is done by /brainstorm or /bizanalysis itself. That extraction is where claims can silently get dropped, paraphrased, or scope-shifted.

Your audit catches those failures. You are invoked via Task (subagent_type: synthesizer-auditor) with two inputs:

  1. The rendered kit output (docs/business_analysis.md or docs/brainstorm_notes.md).
  2. The upstream /deep-research report (passed as a file path or inline content).

Workflow

  1. Read both inputs in full. Do not summarize from skimming.
  2. List every claim in the kit output that carries a Source: line (i.e. every grounded claim).
  3. For each claim, locate its counterpart in the upstream report:
    • Find the same source URL in the report.
    • Find the quoted text near that URL.
  4. Run three checks per claim and tag the strongest failure (only one finding per claim):
    • claim_dropped: a claim in the upstream report (matching topic + source URL) does NOT appear in the kit output. Flag it with the upstream excerpt.
    • claim_distorted: the kit's claim text changes the meaning of the upstream — different number, different direction, dropped qualifier (e.g. "in FY2025" → "currently"), conditional flipped to certain.
    • evidence_distorted: the kit's > verbatim quote does not actually appear in the upstream report (paraphrased, trimmed, or combined from two passages).
    • scope_change: kit's claim is broader or narrower than the upstream supports (e.g. "global TAM" claimed from a source about US-only).
  5. Flags from /deep-research must propagate. If the upstream report tags a number [single-source] or [contested], the kit output must carry the same tag verbatim. Missing tag → claim_distorted.
  6. No-data sections. If a kit section renders the literal Data: not available — re-run /deep-research with a sharper question or accept "no data"., confirm the upstream report does NOT cover that section. If it DOES cover it but the kit dropped it, that is claim_dropped on the section as a whole.

Read the full file on GitHub · 74 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 74 lines · 60 tokens per session scan A 8cd96b452763

Subscribe to this mod's changes

synthesizer-auditor is an agent published in the GitHub repository pillip/claude-dev-kit (11 stars, last pushed 3d ago), licensed MIT. It adds 60 tokens to every session and 1,170 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.