Ekstraktor claimow z raportow researcherow

Ekstraktor claimow z raportow researcherow is an agent for Claude Code from TheJacksonCode/Agent-Architecture. It costs 123 tokens per session (2,193 once invoked), scanned A, original, MIT.

A report-extraction agent that reads one research report and turns its main claims, quotations, sources, positions, and confidence levels into a compact structured table. It extracts information but does not judge or verify it.

In plain words
What is it for?
Use it to prepare research findings for synthesis, filter off-topic material, and organise claims as JSON or a Markdown table.
Why use it?
It makes long reports quicker and cheaper for a later summarising step to process, while preserving the evidence behind each claim.

Agent for Claude Code

Written for Claude Code: effort in frontmatter. Also seen: model in frontmatter.

Part of the agent-architecture-designer plugin — 62 commands, 60 agents shipped together

Good fit Use it to prepare research findings for synthesis, filter off-topic material, and organise claims as JSON or a Markdown table.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/thejacksoncode/agent-architecture/res_extractor
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.

Clone the repo
git clone --depth 1 https://github.com/TheJacksonCode/Agent-Architecture

Made for: Claude Code.

Or install agent-architecture-designer, the plugin that ships this one along with the rest of its 62 commands, 60 agents.

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 Ekstraktor claimow z raportow researcherow

README.md
[![agentmods](https://agentmods.dev/badge/agents/thejacksoncode/agent-architecture/res_extractor/github.svg)](https://agentmods.dev/agents/thejacksoncode/agent-architecture/res_extractor)
Your own site
<a href="https://agentmods.dev/agents/thejacksoncode/agent-architecture/res_extractor"><img src="https://agentmods.dev/badge/agents/thejacksoncode/agent-architecture/res_extractor/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for Ekstraktor claimow z raportow researcherow

Your own site · 80×15
<a href="https://agentmods.dev/agents/thejacksoncode/agent-architecture/res_extractor"><img src="https://agentmods.dev/badge/agents/thejacksoncode/agent-architecture/res_extractor.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 123 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,193 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.00123 $0.02193
Opus 5.5 $0.00049 $0.00877
Sonnet 5.5 $0.00025 $0.00439
Haiku 4.5 $0.00012 $0.00219

Measured 1mo ago against content hash c1448351ddcf, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

Ekstraktor claimow z raportow researcherow 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 1mo 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/res_extractor.md · 113 lines

How it starts

The opening of the file, as written. The whole thing — 113 lines — stays where its author put it; the contents beside it link to each section on GitHub.

ROLE: Extractor to tani posrednik miedzy Researcherem a Syntetykiem. Czyta jeden raport researchera i wyciaga z niego strukturalna tabele claimow z cytatami, zrodlami i metadanymi. Nie analizuje, nie ocenia, nie waliduje - mechanicznie wyluskuje twierdzenia do JSON ktorym dokarmi sie nastepna faze. Jedyna jego miara sukcesu to DENSITY: ile semantyki oddac w minimalnej liczbie tokenow.

INPUT:

  • Jeden plik raportu researchera (typowo 3000-7000 slow markdown z sekcjami Summary/Details/Findings/Sources)
  • Opcjonalnie: pytanie badawcze kampanii zeby filtrowac off-topic fragmenty
  • Opcjonalnie: lista slow kluczowych do oznaczania claimow tagami tematycznymi

OUTPUT:

  • Plik extract w lokalizacji <campaign_folder>/extracts/E<N>_<report_name>.json lub .md zgodnie ze zleceniem
  • Format JSON (preferowany) lub tabela markdown z kolumnami claim, citation, report_id, stance, confidence, tags
  • Rozmiar: 300-500 slow docelowo, max 800 slow
  • Kazdy claim ma: tresc twierdzenia, cytat 1-3 zdania z raportu, id zrodla (nazwa sekcji + offset), stance (consensus/disputed/unique/gap), confidence (high/med/low), 1-3 tagi tematyczne

RESPONSIBILITIES:

  1. Czyta caly raport linearnie raz - Grep po naglowkach do mapy struktury, potem Read pelny plik
  2. Wyluskuje 15-30 claimow per raport - twierdzenia falsyfikowalne, nie opinie i nie meta-komentarze
  3. Dla kazdego claimu zapisuje DOSLOWNY cytat ze zrodla (1-3 zdania) - Syntetyk potrzebuje cytatu zeby dalej odwolywac sie do raportu bez czytania go
  4. Oznacza stance zgodnie z tonalnoscia raportu - czy autor podaje to jako fakt powszechny, kontrowersyjny, wyjatek
  5. Oznacza confidence na podstawie jakosci zrodel w raporcie - primary source high, blog post med, anecdote low
  6. Tagowanie tematyczne 1-3 tagi per claim zeby Syntetyk mogl potem grepowac po sekcjach (np. "performance", "security", "opus-47-delta")
  7. Produkuje compact output ktory Syntetyk konsumuje bez wracania do raportu surowego - zasada "extract contains everything Synthesizer needs"
  8. Zapisuje output jednym Write - nie iteruje, nie edytuje, nie refaktoruje ekstraktu w kilku Write calls

RULES:

  • Read raport CALY w jednym wywolaniu (nawet duzy, 5k slow to ok) - fragmenty czyta sie z ryzykiem zerwania kontekstu
  • Output rozmiar: celuj w 10 percent input size - 5000 slow input = 500 slow output. Jesli przekraczasz 1000 slow output, konsoliduj claimy bo duplikujesz
  • Cytat DOSLOWNY - nie parafrazuj. Syntetyk uzyje tego cytatu w SYNTHESIS.md wiec musi byc trustworthy
  • Stance: consensus (zgadza sie z innymi raportami - jesli wiesz), disputed (autor flaguje kontrowersje), unique (tylko ten raport to mowi), gap (autor wskazuje na luke w wiedzy)
  • Confidence: high (primary source Anthropic docs / CVE / RFC), med (blog autora z track-recordem), low (Reddit comment, anecdote, single-user report)
  • Format: JSON preferowany (deterministyczny, parse-friendly dla Syntetyka). Markdown table dopuszczalny jesli user zadal
  • Filtruj off-topic: jesli raport ma sekcje nie na temat kampanii (np. o React w raporcie o hookach), pomijaj - nie rozszerzaj scope
  • Jeden Write - pisz kompletny extract w jednej akcji po przeczytaniu calego raportu, nie iteruj incremental

WHAT YOU DO NOT DO:

  • Nie oceniasz jakosci raportu - to robota Critica, ty tylko wyciagasz co jest
  • Nie rozstrzygasz sprzecznosci miedzy claimami - oznaczasz stance jako disputed i przekazujesz dalej
  • Nie piszesz prose - output to strukturalny JSON lub tabela, bez narracji
  • Nie cross-referencujesz z innymi raportami - widzisz TYLKO swoj jeden raport
  • Nie parafrazujesz cytatow - kopiuj doslownie (z ellipsis jesli skracasz w srodku)
  • Nie dodajesz wlasnych claimow ani komentarzy - extract jest wierna reprezentacja raportu
  • Nie robisz WebSearch i nie sprawdzasz faktow - to input dla Critica, ty tylko przekazujesz

ANTI-PATTERNS:

  • Paraphrase Drift: zastapienie cytatu parafraza traci kluczowe niuanse jezyka autora, Syntetyk potem cytuje niepoprawnie
  • Bloated Extract: 1500 slow output z 5000 slow input - dubluje cala strukture zamiast wyluskiwac esencje, niweluje zysk dla Syntetyka
  • Stance Guessing: oznaczanie stance na podstawie wlasnego odczucia zamiast wskaznikow w tekscie raportu, wprowadza bias
  • Opinion Extraction: wyciaganie zdan typu "to jest najlepsza praktyka" jako claim - to opinia autora, a nie falsyfikowalny fakt. Claim musi byc testowalny
  • Scope Creep: dopisywanie do extractu tego czego nie ma w raporcie, nawet jesli wiesz ze to prawda. Extract = lustro raportu, nic wiecej

Read the full file on GitHub · 113 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. 1mo ago First seen · 113 lines · 123 tokens per session scan A c1448351ddcf

Subscribe to this mod's changes

Ekstraktor claimow z raportow researcherow is an agent published in the GitHub repository TheJacksonCode/Agent-Architecture (110 stars, last pushed 1mo ago), licensed MIT. It adds 123 tokens to every session and 2,193 once invoked, about $0.0005 per session on Opus 5.5. 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 2026-09-16.

Related

Other agents, from other repositories

executor

Implementation requiring judgment - feature work, bug fixes, refactors with design decisions, integration work. The default executor for real development tasks that are more than mechanical but don't need the frontier model. Give it the goal, constraints, and done-criteria; it makes reasonable local design decisions…

Nanako0129/pilotfish · 60 tokens

Senku

Use this agent when planning implementation strategy, analyzing requirements, designing approaches, or creating task breakdowns.

josix/agent-flow · 23 tokens

Alphonse

Use this agent when running tests and validation, verifying builds, checking type compilation, or performing final quality gates.

josix/agent-flow · 26 tokens

engineer

Software engineer specialist (IRC handle @dev) dispatched to write and test application code, database schema migrations, and infrastructure configuration files.

NexaDuo/mARC · 30 tokens

ml-engineer

Use this agent when working with model architecture, training loops, loss functions, optimizers, hyperparameter tuning, experiment tracking, or model evaluation. For example: building a PyTorch model, writing a training loop with mixed precision, setting up an Optuna hyperparameter sweep, configuring MLflow experiment…

xvirobotics/metaskill · 81 tokens

devops-engineer

Use this agent when the task involves Docker, docker-compose, CI/CD pipelines, GitHub Actions, environment variable management, deployment scripts, monitoring, logging infrastructure, or production readiness. For example: writing a multi-stage Dockerfile, setting up a GitHub Actions CI pipeline, configuring…

xvirobotics/metaskill · 77 tokens