RES SaaS Recon

Turn browser-observed SaaS behavior into an evidence-backed rebuild kit.

Source repo
alextnetto/res
Stars
★ 313
Last updated
1mo ago
License
MIT
Primary language
JavaScript

At a glance

How it runs
Agent plugin / skillCLI
Works with
Portable with changesClaude Code
Cost
Free software; you pay for model usage
Setup effort
Medium · a few setup steps
You'll need
Node.js 22 or laterClaude in Chrome, Playwright MCP or Chrome DevTools MCPShell / CLINetwork accessLocal filesystemMCP Server
Typical use
A product team rebuilding a CRM for a target market can document sales flows, permissions, and rules in a trial account.
Not a fit if
  • Teams that need deep analysis of desktop or mobile app internals
  • Teams without an authorized test account or sandbox
Website
alextnetto.github.io ↗ set on GitHub
Maintenance
Commits in the last 6 monthslast commit 2026-10-17latest release v0.4.0 (2026-10-17)+2.4 ★/day over 7 days
Source review
76/100 · Good

What does this agent do, and when should you use it?

RES is an agent skill for Claude Code and other agents that can read skills from disk and operate a browser, packaged with Markdown templates and dependency-free Node.js scripts. It studies an authorized web product in stages: read public documentation, check the environment, map screens, then investigate features one at a time. Its knowledge base records screen states, business rules, data models, API shapes, design tokens, and launch barriers. At close, scripts can produce a self-contained REPORT.html for people and a handoff zip containing BUILD.md and the knowledge base. Safety levels, a change ledger, and teardown steps define the write boundary.

Install RES from the Claude Code plugin marketplace, or copy the reverse-engineering-saas skill folder into a compatible agent's skills directory. The agent works in a trial account in a separate browser profile, an empty workspace, or an authorized sandbox; it uses web fetch for public documentation and browser tools plus page JavaScript to inspect screens, routes, states, interactions, network activity, and design tokens. Findings are written to files with [obs], [net], [doc], or [inf] evidence labels, and writes are recorded in sandbox/ledger.csv. At the end, run check-kb.mjs to check the knowledge base, then build-report.mjs and pack-kb.mjs to create the report and handoff package.

  1. A product team rebuilding a CRM for a target market can document sales flows, permissions, and rules in a trial account.
  2. An engineering team estimating a SaaS replacement can map screens, data relations, API shapes, and parity tests first.
  3. A product manager preparing a launch plan can identify integration approvals, certifications, licenses, and distribution barriers.
  4. A design team matching an existing product can inventory screen states, components, microcopy, accessibility behavior, and design tokens.
  5. A team doing research across multiple agent sessions can keep the knowledge base on disk and track human follow-ups in STATUS.md.

How do you install or deploy this agent?

You need Node.js 22 or later and an agent/browser setup that can drive a browser and run JavaScript in the page. For Claude Code:

How do you use this agent?

For the safest setup, create a trial account you control, log in using a separate Chrome profile, and connect the agent's browser tool to that profile. Then give the agent a study request in a Claude Code session, for example:

What are this agent's strengths and limitations?

Pros
  • Combines screens, rules, data models, API shapes, design tokens, and production barriers in one evidence-backed knowledge base.
  • Its L1–L4 safety model records writes, checks undo paths, and defines end-of-study restoration.
  • Includes init-kb.mjs, check-kb.mjs, build-report.mjs, and pack-kb.mjs for creating, checking, reporting, and packaging a knowledge base.
  • Includes end-to-end ClickUp and Pipedrive study reports and knowledge bases.
Limitations
  • Research depends on an agent and browser tool that can drive a browser and run page JavaScript; the README names Claude in Chrome, Playwright MCP, and Chrome DevTools MCP.
  • Requires Node.js 22 or later, a browser session, and an authorized product environment for a full study.
  • The README says it is not well suited to desktop or mobile apps because it needs the DOM and network.
  • Whether copying a product is legal depends on its terms and jurisdiction; RES flags relevant terms but does not provide legal advice.

How does this agent compare with similar options?

The README contrasts RES with REA: REA targets binaries, Android, Electron, network captures, and websites to show what a product is built with; RES documents how it works for users and what it takes to rebuild and launch it.

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
RES SaaS Recon This agent 76 · Good Agent plugin / skillFree + model costs ★ 313 1mo ago JavaScript Claude Code
assay 81 · Good CLIFree ★ 110 15d ago Python Codex · Claude Code
Stagehand: The AI Browser Automation Framework 68 · Some gaps Library / SDKFree + model costs ★ 26k today TypeScript —
Playwright Browser Automation MCP 65 · Some gaps MCP serverFree ★ 40k 2d ago TypeScript Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Good
76/ 100 5-point scale 3.8 / 5
Trust 23/29
Reliability 8/14
Adaptability 15/18
Convention 14/18
Effectiveness 10/13
Verifiability 6/8
Why each dimension lost points
Trust23 / 29 · 4.0/5

Evidence shows a clear tiered permission model (L1-L4), prohibitions on sending messages/invites/mentions/billing/account creation/password entry, and pre-write checks of location, audience, undo path and ledger — strong least-privilege design, so least_privilege = 3. However, user confirmation relies mainly on inferring intent from the initial request ('test everything' counts as write authorization) rather than explicit per-write confirmation, so user_confirmation is deducted to 2. On data flow, README states account details (sandbox/, log/) stay out of the zip with only a de-identified ledger, but page scripts may read page values and SECURITY.md acknowledges a data-leak risk class, so data_flow_transparency and sensitive_data_handling are each deducted to 2. Dependencies: the skill itself has none, only a pinned playwright devDependency, but no vulnerability scanning or lockfile evidence is shown, so dependency_security = 2. External effects and rollback have ledger.csv, restore/remove/list-for-resolution and teardown flows, so external_effects and rollback = 3. Source attribution has an [obs]/[net]/[doc]/[inf] tagging system, but no concrete license tracking for reused third-party assets, so source_attribution = 2.

Reliability8 / 14 · 2.9/5

SKILL.md and README phase tables, environment levels and evidence tags are broadly consistent, so self_consistency = 2 (some content is truncated, so full consistency cannot be confirmed). Dependency availability: Node 22+, pinned playwright 1.64.0, CI on 22/24, so dependency_availability = 2. Failure messages: only check-kb and the STATUS.md 'Needs human' mechanism are visible; no concrete error-message design or detailed failure-recovery docs, so failure_messages = 1.

Adaptability15 / 18 · 4.2/5

Audience and scenarios are well covered (team rebuild, competitor, coding-agent handoff) with two end-to-end examples (ClickUp/Pipedrive), so audience_and_scenarios = 3. Capability boundaries are clear: not for source-available code, native binaries, Electron internals, or single-feature questions, so capability_boundaries = 3. Trigger precision: the description is long and lists many product names, risking over-triggering, so trigger_precision = 2. Environment fit supports Claude in Chrome, Playwright MCP, Chrome DevTools MCP and separate profiles, but explicitly works poorly for desktop/mobile apps, so environment_fit = 2.

Convention14 / 18 · 3.9/5

Information architecture is strong: phases, directory layout, ID system, templates, so information_architecture = 3. Install notes are complete (marketplace, manual copy, direct scripts, update commands), so install_notes = 3. Naming stability: version 0.4.0 with 0.3.0 fixing gaps found by the Pipedrive study indicates evolving structure, so naming_stability = 2. Examples and FAQ are rich, so examples_and_faq = 3. Known limitations appear in FAQ and Status but are not systematically enumerated, so known_limitations = 2. License is clearly MIT, so license = 3. A version number exists but there is no standalone CHANGELOG, so versioning_changelog = 2. Maintenance responsibility: publisher is unverified by FollowAgents, SECURITY.md only says the latest release gets fixes, and there is no explicit maintainer commitment or SLA, so maintenance_responsibility = 1.

Effectiveness10 / 13 · 3.8/5

Output usability is strong: self-contained REPORT.html, zip with BUILD.md, rules convertible to parity tests, so output_usability = 3. Marginal value: it adds structured KB, evidence tags and barrier analysis over a generic browser agent, but the core is still prompt orchestration, so marginal_value = 2. Cost-benefit: ~130 tokens resident and ~4,600 at run time with rich output, but it depends on a real browser and manual safe-environment setup, so cost_benefit = 2.

Verifiability6 / 8 · 3.8/5

claim_traceability: every statement is required to carry [obs]/[net]/[doc]/[inf] evidence, and rule/barrier examples cite evidence, so 2 (no independent validation of the evidence itself is shown). cross_source_corroboration: supports public docs, network capture and page observation as multiple sources, but does not mandate cross-verification, so 2. fact_inference_separation: explicitly separates [inf] inferences and open questions, so fact_inference_separation = 3.

Risks and how to mitigate them
  • The skill drives a real browser on real accounts; users must ensure authorized use and prefer a trial account in a separate profile.
  • Write authorization relies on inferring intent from the initial request (e.g. 'test everything'), risking misread authorization; explicit confirmation before key writes is advisable.
  • Page scripts may return sensitive page values (names, emails, IDs, tokens); SECURITY.md acknowledges this leak risk, so script return values should be reviewed.
  • Publisher identity is unverified by FollowAgents and maintenance commitment is unclear; only the latest release receives fixes.
  • Support for desktop/mobile apps is poor, and it depends on Node 22+ and specific browser tooling.
Evidence confidence: Low Reviewed Nov 10, 2026 Reviewed revision 1d05d9c4a9f9
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/res.svg)](https://followagents.com/en/agents/res)
<a href="https://followagents.com/en/agents/res"><img src="https://followagents.com/badge/res.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v0.4.0 RES 0.4.0
    Release notes excerpt
    - REPORT.html redesigned as a UX study a person can read and navigate: IBM Plex type, a light and a dark theme with a toggle, a one-line answer with the key findings, stat tiles, section chips, and screenshots in framed figures that enlarge on click. - Screens: anatomy screenshot beside a numbered legend linked to the components, a gallery of the screen's evidence, the states seen and the states not reached. - Components: one card per component with a screenshot, variants and states as pills (solid when seen, dashed when not), and behavior notes. …
    View on GitHub ↗
  2. v0.3.1 RES 0.3.1
    Release notes excerpt
    - REPORT.html redesigned for reading: a header with the product and study status, a notice that the page is for people and that AI agents should read BUILD.md (also as a comment at the top of the source), numbered sections in reading order (overview, product, screens, flows, rules, data, API, design, build plan, then what to prepare before launch), a coverage bar, rules and barriers as cards, IDs as badges.
    View on GitHub ↗
  3. v0.3.0 RES 0.3.0
    Release notes excerpt
    First end-to-end studies on real products, published in examples/ (ClickUp at L2, Pipedrive at L3, each with REPORT.html), and fixes for the gaps the Pipedrive study found. - Handoff: pack-kb.mjs keeps sandbox/ledger.csv in the zip with object IDs and locations blanked and URLs replaced by [url]. An unzipped handoff now passes check-kb --closing, and L-### references in the KB resolve. …
    View on GitHub ↗

FAQ

Will it change production data?
It writes only inside a sandbox you allowed. Each change is recorded in sandbox/ledger.csv and is restored, removed, or listed for you to resolve at the end. Without write permission, it only observes.
Does the agent need my password?
No. You log in yourself, preferably to a trial account in a separate browser profile. The agent never types passwords or creates accounts.
Can I use it to study desktop or mobile apps?
Not well. RES needs the DOM and network; the README points to REA for binaries and Electron internals.
Can it run without interrupting me?
Yes. Tell the agent not to ask questions; it records human follow-ups in STATUS.md, continues other work, and restores its changes before stopping.
Is RES SaaS Recon free?
The software is free, but RES SaaS Recon calls an AI model, and you pay that model provider for usage (or run a local model).
Is RES SaaS Recon open source?
Yes. The source code is on GitHub (alextnetto/res) under the MIT licence.
Is RES SaaS Recon safe to use?
FollowAgents' FARS review (Nov 10, 2026) scored RES SaaS Recon 23/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is RES SaaS Recon still maintained?
Development has slowed: the latest commit on GitHub is from 2026-10-17, within the last six months but not the last 30 days. The latest release is v0.4.0 (2026-10-17).
Who makes RES SaaS Recon?
RES SaaS Recon is published on GitHub by the account alextnetto. FollowAgents has not verified which person or company is behind that account.
What is RES SaaS Recon's official website?
The project's GitHub repository lists https://alextnetto.github.io/res/examples/pipedrive/REPORT.html as its website. That link is set by the repository owner; the source code itself is on GitHub (alextnetto/res).
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts