Autonomous Agents MCP Server Knowledge Pack

Autonomous Agents

Universal methodology for autonomous agents — four role playbooks (SE, QA, SD, PM) plus operational workflows for running agent fleets. App-agnostic: works with any web application or codebase.

Role	Focus	Purpose
SE	Software Engineer	Pick issues, write code, create PRs, iterate on review feedback
QA	Quality Assurance	Test web applications, collect evidence, document bugs
SD	Skill Discovery	Explore web apps, document modules, generate knowledge packs
PM	Product Management	Sprint planning, issue triage, progress reporting

Each role file (_roles/se.md, _roles/qa.md, _roles/sd.md, _roles/pm.md) contains the complete methodology for that role type. Domain-specific knowledge packs extend these universal roles with app-specific context.

The ops/ module contains the shared workflow catalog for dispatching and operating agent fleets at scale.

---

SD Methodology — Skill Discovery & Web App Documentation

A universal methodology that teaches autonomous agents how to explore, document, and test any web application through browser automation. It produces knowledge packs — structured domain knowledge that SE and QA agents depend on.

What SD Does

The SD role is a meta capability: it doesn't automate a specific app — it teaches the fleet HOW to discover and document any web app. Point an SD session at any URL, and it produces a complete knowledge pack for that application.

Three Capabilities

Capability	Role	Output
Skill Discovery	SD	_skill.md files documenting every module: selectors, data model, states, tasks
QA Testing	QA	Test playbooks per module, evidence collection, bug documentation
Role Generation	SD	_roles/qa.md and _roles/se.md files derived from skill documentation

How It Works

1. Point SD at any web app URL
2. SD explores navigation, discovers modules, estimates features
3. SD progressively documents each module (selectors, data model, states, tasks)
4. At 75% coverage → SD generates QA and SE role files
5. QA uses generated playbooks to test the app
6. SD polishes docs based on QA findings

Skill Pack Output Structure

SD produces a complete skill pack for the target app:

{target-domain}/
+-- skill-pack.json              # Pack manifest
+-- _skill.md                    # Root: app overview + Module Inventory table
+-- _roles/
|   +-- qa.md                    # Root QA methodology
|   +-- se.md                    # Root SE architecture
|   +-- sd.md                    # App-specific discovery notes
+-- {module}/
|   +-- _skill.md                # Module docs (9 standard sections)
|   +-- _roles/
|   |   +-- qa.md                # Per-module test playbook
|   |   +-- se.md                # Per-module code ownership
|   +-- *.yaml                   # Browser automation workflows

_skill.md Standard Sections

Every module _skill.md must contain these 9 sections:

#	Section	Content
1	What This Is	One-paragraph description of the module's purpose
2	URL Patterns	All URL patterns for list/detail/sub-views
3	Page Structure	Layout description: columns, regions, component hierarchy
4	Key Elements	Table: Element / Selector / Notes — every interactive element
5	Data Model	Table: Field / Type / Values / Default — entity schema from the UI
6	States	Named page states: Default, Filtered, Empty, Editing, Modal Open, Error
7	Common Tasks	Numbered step-by-step for each user operation
8	Tips	Non-obvious behaviors, shortcuts, defaults
9	Gotchas	Automation pitfalls: timing, portals, selector instability

YAML frontmatter required: name, domain, path, parent, tags, learned, last_verified, confidence.

Module Inventory Table

The root _skill.md tracks all modules and their documentation coverage:

| Module | Features | `_skill.md` | `qa.md` | `se.md` | Brief Description |
|--------|:--------:|:-----------:|:-------:|:-------:|-------------------|
| login  | ~3       | 90%         | 85%     | —       | Authentication flow |
| home   | ~8       | 50%         | —       | —       | Main dashboard |
| users  | ~15      | 1%          | —       | —       | Discovered, not yet documented |

Progressive Coverage Model

Coverage percentages track documentation completeness:

_skill.md Coverage

% Range	What's Done
1%	Module discovered: name, URL, one-line description
10-25%	What This Is + URL Patterns + Page Structure (skeleton)
25-50%	+ Key Elements table (major interactive elements)
50-75%	+ Data Model + States + Common Tasks
75-90%	+ Tips + Gotchas + all element selectors verified
90-100%	All 9 sections complete, selectors verified against live app

Role File Coverage

Role files are generated when _skill.md reaches 75%, starting at 45%:

% Range	What's Done
45%	Generated from template: structure + test phases matching Common Tasks
45-70%	+ Detailed test criteria, + specific pass/fail conditions
70-90%	+ Automation Tips with module-specific patterns
90-100%	+ Edge case tests, cross-module integration tests

Browser Tools Reference

Tool	Purpose	When to Use
navigate(url)	Go to page	Every module visit
snapshot(includeContent=true)	Capture DOM tree + content	Primary discovery tool
screenshot()	Visual capture	Evidence, layout verification
click(ref=N)	Click element by ref	Interaction testing
type(ref=N, text)	Type into input	Form testing
scroll(direction, amount)	Scroll page	Below-fold content
get_browser_status()	Check connection	When tools start failing

Getting Started

1. Install this knowledge pack: npx sidebutton install agents
2. Start SideButton: npx sidebutton
3. Connect browser (Chrome extension)
4. Navigate to target app
5. Run SD with target URL — it handles everything from there

Requirements

• SideButton server running (npx sidebutton)
• Chrome with SideButton extension connected
• Target web application accessible in browser
• Claude Code or compatible LLM orchestrator