copilot-operator 1.0.3

Autonomous orchestration loop for GitHub Copilot in VS Code — drives multi-iteration coding sessions with self-correction, learning, and intelligent strategy.

Copilot Operator

Autonomous meta-agent that drives GitHub Copilot Chat to complete coding tasks end-to-end.

Copilot Operator doesn't write code itself — it controls GitHub Copilot Chat via the VS Code CLI, evaluates results, detects stuck loops, and adapts its strategy until the task is done.

---

Features

Feature	Description
Operator Loop	Send prompt → parse response → evaluate score → decide continue/stop
Intelligence Engine	Trend analysis, loop detection, adaptive strategy hints
Goal Decomposition	Auto-classify goals (bug/feature/refactor/docs/audit/stabilize) and build milestone plans. LLM-powered when available.
Validation	Run real test/lint/build commands between iterations
Meta-Learning	Detect failure patterns, generate prompt guardrails for future runs
Adversarial Review	Coder + Critic self-review before accepting results
Snapshot & Rollback	Git stash snapshots each iteration; auto-rollback on score regression
LLM Brain	Connect to OpenAI, Anthropic, Gemini, or local models for deeper analysis
Repo Map	AST + regex codebase index injected into every prompt (15 languages) — like Aider
GitHub Integration	Auto-fetch issues, create PRs — all via REST API (stdlib only)
Cross-Repo Brain	Share learnings across repositories via ~/.copilot-operator/shared-brain/
Adaptive Guardrails	Static + dynamic guardrails that evolve based on run history
Benchmark Engine	Run and score operator cases against keyword expectations
Live Mode	Colour-coded real-time iteration progress in the terminal (--live)
Dry-Run Mode	Generate prompts without VS Code interaction — safe for testing
Error Recovery	Retry on transient errors, stop on consecutive failures
Circuit Breaker	Rate-limit protection for LLM and GitHub API calls

---

Quick Start

Requirements

• Python 3.10+
• VS Code with GitHub Copilot extension
• code CLI available in PATH

Install

pip install copilot-operator

Or from source:

git clone https://github.com/copilot-operator/copilot-operator
cd copilot-operator
pip install -e .

Setup

# Scaffold workspace config
copilot-operator init

# Pre-flight checks
copilot-operator doctor

Run

# Autonomous run with a goal
copilot-operator run --goal "Fix the login timeout bug in auth.py"

# Live colour-coded progress
copilot-operator run --goal "Add pagination to API" --live

# Dry-run: see the generated prompt without executing
copilot-operator run --goal "Add pagination to API" --dry-run

# Resume a stopped or blocked run
copilot-operator resume

# Fix a specific GitHub issue
copilot-operator fix-issue --issue 42 --repo owner/repo

# Watch progress live
copilot-operator watch

# Run benchmark cases
copilot-operator benchmark --file benchmark.json

---

LLM Brain (Optional)

Set environment variables or create .env in your workspace:

# OpenAI
COPILOT_OPERATOR_LLM_PROVIDER=openai
OPENAI_API_KEY=sk-...

# Anthropic
COPILOT_OPERATOR_LLM_PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-...

# Gemini
COPILOT_OPERATOR_LLM_PROVIDER=gemini
GEMINI_API_KEY=...

# Local (Ollama, LM Studio, etc.)
COPILOT_OPERATOR_LLM_PROVIDER=local

Check status:

copilot-operator brain
copilot-operator brain --test "What is 2+2?"

---

Benchmark

Create a benchmark.json to measure operator quality:

{
  "name": "My project benchmark",
  "cases": [
    {
      "id": "fix-auth-bug",
      "goal": "Fix the login bug where tokens expire too early",
      "goal_profile": "bug",
      "expected_keywords": ["token", "expiry", "authentication"]
    },
    {
      "id": "add-readme-docs",
      "goal": "Update README with installation and usage sections",
      "goal_profile": "docs",
      "expected_keywords": ["README", "installation", "usage"]
    }
  ]
}

# Human-readable report
copilot-operator benchmark --file benchmark.json

# Machine-readable JSON output
copilot-operator benchmark --file benchmark.json --json

---

Architecture

┌─────────────────────────────────────────────────────┐
│                   CLI (cli.py)                       │
├──────────┬──────────────────────────────────────────┤
│          │         CopilotOperator (operator.py)     │
│          │  ┌──────────────┐  ┌──────────────────┐  │
│  Config  │  │  run() loop  │  │  _decide() logic │  │
│   YAML   │  │  ↓ prompt    │  │  score gates     │  │
│  + .env  │  │  ↓ send      │  │  blocker checks  │  │
│          │  │  ↓ parse      │  │  replan triggers │  │
│          │  │  ↓ validate   │  │  rollback logic  │  │
│          │  │  ↓ decide     │  │  critic checks   │  │
│          │  └──────────────┘  └──────────────────┘  │
├──────────┼──────────────────────────────────────────┤
│  Intelligence Layer                                  │
│  ┌────────────┐ ┌──────────┐ ┌──────────────────┐  │
│  │ reasoning  │ │  brain   │ │  meta_learner    │  │
│  │ (trends,   │ │ (project │ │  (pattern detect,│  │
│  │  loops)    │ │  history)│ │   prompt rules)  │  │
│  └────────────┘ └──────────┘ └──────────────────┘  │
├──────────┼──────────────────────────────────────────┤
│  External Integrations                               │
│  ┌────────────┐ ┌──────────┐ ┌──────────────────┐  │
│  │ llm_brain  │ │ snapshot │ │ github_integration│  │
│  │ (4 LLM     │ │ (git     │ │ (issues, PRs)    │  │
│  │  providers)│ │  stash)  │ │                   │  │
│  └────────────┘ └──────────┘ └──────────────────┘  │
├──────────┼──────────────────────────────────────────┤
│  VS Code Bridge                                      │
│  ┌──────────────┐ ┌────────────┐ ┌──────────────┐  │
│  │ vscode_chat  │ │ session    │ │ validation   │  │
│  │ (CLI bridge) │ │ store      │ │ (subprocess) │  │
│  └──────────────┘ └────────────┘ └──────────────┘  │
└─────────────────────────────────────────────────────┘

---

Configuration

copilot-operator.yml

workspace: .
mode: agent
goalProfile: default          # bug | feature | refactor | audit | docs | stabilize
maxIterations: 6
targetScore: 85
sessionTimeoutSeconds: 900

validation:
  - name: tests
    command: npm test
    required: true
  - name: lint
    command: npm run lint
    required: false

llm:
  provider: openai            # openai | anthropic | gemini | local
  model: gpt-4o

.copilot-operator/repo-profile.yml

repoName: my-project
summary: A Node.js REST API
standards:
  - Use TypeScript strict mode
  - All functions must have tests
priorities:
  - Test coverage > 80%
protectedPaths:
  - database/migrations/

---

CLI Reference

Command	Description
copilot-operator doctor	Pre-flight checks (VS Code, config, validations)
copilot-operator init	Scaffold config files and documentation
copilot-operator run --goal "..."	Start a new autonomous run
copilot-operator run --goal "..." --live	Run with colour-coded real-time progress
copilot-operator run --goal "..." --dry-run	Generate prompt without executing
copilot-operator resume	Resume from last checkpoint
copilot-operator status	Show current run state
copilot-operator plan	Show the current milestone plan
copilot-operator focus	Show what the operator is working on
copilot-operator watch	Live-poll run progress
copilot-operator brain	Show LLM brain status
copilot-operator brain --test "..."	Test LLM brain with a prompt
copilot-operator fix-issue --issue N --repo owner/repo	Fetch GitHub issue and run operator to fix it
copilot-operator benchmark --file bench.json	Run benchmark cases and score results
copilot-operator cleanup	Remove old run logs
copilot-operator version	Show version
copilot-operator -V	Short version flag

---

How It Works

1. You provide a goal — e.g., "Fix issue #42" or "Add pagination to the API"
2. Operator classifies the goal — determines it's a bug fix, feature, refactor, etc.
3. Operator builds a milestone plan — heuristic-based or LLM-powered
4. For each iteration:
   • Takes a git stash snapshot
   • Runs pre-validation (tests, lint)
   • Builds a rich prompt with context, guardrails, and intelligence
   • Sends the prompt to Copilot Chat via code chat --mode agent
   • Waits for and parses the response
   • Runs post-validation
   • Evaluates: score, blockers, validation results, trend
   • Decides: continue, replan, rollback, or stop
5. After the run:
   • Meta-learner extracts failure patterns as rules for future runs
   • Cross-repo brain exports learnings
   • Results saved for project brain analysis

---

Project Structure

copilot_operator/
├── operator.py            # Main orchestration loop
├── vscode_chat.py         # VS Code CLI bridge (subprocess)
├── session_store.py       # VS Code session file parser
├── validation.py          # Test/lint/build command execution
├── config.py              # YAML config + .env loading
├── cli.py                 # CLI with 15+ subcommands
├── prompts.py             # Prompt templates + response parsing
├── planner.py             # Plan parsing, merging, rendering
├── reasoning.py           # Trend analysis, loop detection
├── brain.py               # Project history analysis
├── goal_decomposer.py     # Goal classification + LLM decomposition
├── repo_map.py            # AST + regex codebase map (15 languages)
├── benchmark.py           # Benchmark runner + scoring engine
├── terminal.py            # ANSI colour helpers (NO_COLOR compliant)
├── scheduler.py           # Multi-session orchestration
├── repo_ops.py            # Git operations (branch, commit, diff)
├── meta_learner.py        # Pattern detection + rule learning
├── adversarial.py         # Coder + Critic review
├── llm_brain.py           # Multi-provider LLM (circuit breaker)
├── github_integration.py  # GitHub REST API (retry + rate limit)
├── snapshot.py            # Git stash snapshots + rollback
├── cross_repo_brain.py    # Shared knowledge across repos
├── intention_guard.py     # Static + adaptive guardrails
├── repo_inspector.py      # Workspace ecosystem detection
├── bootstrap.py           # Workspace scaffolding
├── logging_config.py      # Structured logging
└── py.typed               # PEP 561 type marker

---

License

MIT

Docs index

Read these first if you want a working rollout plan:

• docs/COPILOT_OPERATOR_MASTER_PLAN.md: the long-range architecture and phase roadmap
• docs/COPILOT_OPERATOR_CHECKLIST.md: the actionable checklist, with current items marked done or pending
• docs/COPILOT_OPERATOR_RUNBOOK.md: day-0 setup, operating steps, and unblock flow
• docs/COPILOT_OPERATOR_BACKLOG.md: ticket-ready engineering backlog
• docs/COPILOT_OPERATOR_GOAL_TEMPLATES.md: reusable goal templates for real runs
• docs/operator/: seeded repo-brain files attached to the operator profile

Validation

npm test
npm run operator:doctor
npm run supervisor:doctor