Grunt

An ultra-intelligent, daily-generational, self-evolving system for eliminating human grunt work.

---

What is Grunt?

Grunt is an open source platform that collects, ingests, wires, and integrates agent persona/profile templates using a scientific method that self-iterates toward near-perfection for eliminating units of human grunt work.

Each day, Grunt:

1. Researches the internet to grow its definition of human grunt work
2. Curates and scores agent persona templates against a scientific framework
3. Self-iterates — the weakest templates are refined, the strongest are promoted
4. Reports progress on the key metric: units of grunt work eliminated per cycle

---

Quick Start

# Start Grunt
grunt work

---

Architecture

grunt/
├── agents/                        # Agent persona/profile templates
│   ├── TEMPLATE.md                # Template schema
│   ├── standup-summarizer.md      # ⭐ Gold (0.81)
│   ├── code-reviewer.md           # ⭐ Gold (0.82)
│   ├── release-notes-writer.md    # ⭐ Gold (0.83)
│   ├── pr-description-writer.md   # ✅ Pass (0.77)
│   ├── documentation-updater.md   # ✅ Pass (0.72)
│   ├── bug-report-triager.md     # ✅ Pass (0.70)
│   └── grunt-ceo/                 # Grunt CEO agent (Paperclip)
├── docs/                          # Scientific method, scoring, roadmap
│   ├── VISION.md
│   ├── SCORING.md
│   └── ROADMAP.md
├── scripts/                       # Daily evolution scripts (v0.2)
└── grunt                          # Startup command (grunt work)

Daily Generation Cycle

Internet Research → Grunt Work Identification → Agent Persona Scoring
       ↓                      ↓                         ↓
  Trend Ingestion     Task Taxonomy Update      Template Refinement
                                                         ↓
                                               Self-Iteration Report

---

Agent Persona Templates

Grunt templates live in agents/. Each template defines:

• Persona — name, role, capabilities, tone
• Score — effectiveness rating (0.0–1.0) against the scientific framework
• Lineage — parent templates, iteration history
• Grunt Units — units of human grunt work this persona eliminates per cycle

Template Leaderboard

Template	Version	Composite	Tier	Grunt Work Eliminated
Release Notes Writer	v2	0.83	⭐ Gold	Changelog & release notes from git history
Code Reviewer	v2	0.82	⭐ Gold	Automated PR review for bugs, style, security
Standup Summarizer	v1	0.81	⭐ Gold	Async standup digests — no meetings needed
PR Description Writer	v1	0.77	✅ Pass	PR descriptions from diffs & commit history
Documentation Updater	v1	0.72	✅ Pass	Keeps docs in sync with code changes
Bug Report Triager	v1	0.70	✅ Pass	Auto-categorize, prioritize, and route bug reports

Scoring tiers: ⭐ Gold (≥ 0.80) • ✅ Pass (0.60–0.79) • ⚠️ Review (0.40–0.59) • ❌ Fail (< 0.40)

---

Scientific Scoring Framework

Grunt uses a multi-dimensional scoring framework to measure effectiveness:

Dimension	Weight	Description
Grunt Units Eliminated	40%	Tasks completed per cycle
Autonomy Rate	25%	Tasks completed without human intervention
Error Rate	20%	Inverse of errors per 100 tasks
Iteration Delta	15%	Improvement vs. prior generation

See docs/SCORING.md for full specification.

---

Roadmap

Milestone	Target	Status
v0.1 — Foundation	Q2 2026	🚧 In Progress
v0.2 — Daily Cycle	Q3 2026	Planned
v0.3 — Community Templates	Q4 2026	Planned
v1.0 — Self-Evolving Production	2027	Vision

---

Contributing

See CONTRIBUTING.md and docs/.

---

License

MIT — see LICENSE.