autonomous-dev — A Development Harness for Claude Code

A harness that wraps Claude Code with enforcement, specialist agents, and alignment gates to deliver consistent, production-grade software engineering outcomes.

A harness is the software and structure that wraps an AI model to keep it on track — the prompts, tools, feedback loops, constraints, and validation that turn a capable but undisciplined model into a reliable system. Without a harness, the model is a wild horse with raw power but no direction. With one, that power is controlled, directed, and accountable.

What Is This?

The problem: Claude Code is brilliant at execution but unreliable at process. It skips tests, declares "good enough" on failing code, bypasses security reviews, and drifts from your project's intent. Not out of malice — it's trained to complete tasks, not follow engineering discipline. Prompt-level instructions ("please run tests") get ignored under context pressure.

The solution: autonomous-dev is a harness that enforces the full software development lifecycle:

You define your project in a file called PROJECT.md - your goals, what's in/out of scope, constraints, architecture
You track work in GitHub Issues - each feature or bug is an issue
You run one command - /implement "#72"
Claude executes a full development pipeline - research, planning, acceptance tests, implementation, code review, security scan, documentation
Claude stays aligned - if a feature doesn't fit your PROJECT.md scope, it's blocked before any code is written

In short: You define the rules once. The harness enforces them for every feature.

What Does It Actually Do?

When you type /implement "#72", Claude coordinates specialist agents through a 15-step pipeline (the default full mode). The step numbers below are the real STEP N/15 ids you'll see in the run output and logs:

Step	Stage	What It Does
1–2	pre-flight + alignment	Blocks pre-staged files; checks the feature against PROJECT.md goals/scope. Out of scope → blocked before any code.
3–4	researcher (×2, parallel)	`researcher-local` (Haiku, codebase) + `researcher` (Sonnet, web) find patterns, best practices, existing solutions
5	planner (Opus)	Designs the architecture and a file-by-file implementation plan
5.5	plan-critic (Sonnet)	Adversarial single-pass review across 4 axes (assumption audit, existing-solution search, minimalism, operational-integration test). Verdict PROCEED / REVISE / BLOCKED; REVISE re-invokes the planner. Plus structural validation of the plan.
6	acceptance tests	Writes acceptance tests that define what "done" means (acceptance-first default)
8	implementer (Opus) + test gate	Writes code + unit tests. HARD GATES: 0 test failures, no stubs, bug-fix regression test required, planned-vs-implemented file alignment (blocks >50% divergence)
8.5	spec-validator (Opus)	Spec-blind behavioral validation — writes tests from acceptance criteria without seeing the implementation, then validates against it
9–9.5	registration + agent-count gates	Verifies new hooks are registered; blocks if required pipeline agents haven't run
10	reviewer + security-auditor + doc-master	Run in parallel for low-risk changes; sequential (reviewer → security) on security-sensitive paths. Review across 6 dimensions; OWASP scan; doc-drift check
11	remediation gate	Blocking findings loop back to the implementer, max 2 cycles, then block-and-file
12.5	continuous-improvement-analyst	Examines the session for drift, bypasses, gaming — dispatched before commit (#1211)
13–15	git + doc-congruence + cleanup	Commit (blocked until agents complete), push, close issue; documentation-congruence HARD GATE; state cleanup

One command. Full software development lifecycle. Aligned to your project.

Why Use This?

Without a harness:

You ask Claude to build a feature
Claude makes assumptions about your architecture
It skips tests when context gets long (context anxiety)
It approves its own work as "good enough" (poor self-evaluation)
You review the code, find issues, ask for fixes
Repeat until it's acceptable
Hope you didn't introduce security issues

With autonomous-dev:

You define your project once (PROJECT.md)
You create a GitHub issue for each feature
You run /implement "#X"
The harness enforces every step — research, planning, testing, implementation, adversarial review, security audit, documentation
You review a complete, tested, documented, security-scanned result

The difference: Claude stops guessing. The harness keeps it honest.

Four-Layer Harness Architecture

autonomous-dev enforces process through four layers, each addressing a different failure mode — enforce now, add intelligence, learn from the session, then evolve:

1. Hooks (deterministic enforcement) — Run on every tool call, commit, and prompt. Hard gates that can't be argued with: tests must pass with 0 failures before code review starts, no stubs or placeholders allowed, security scan is mandatory, documentation must stay in sync. Functional infrastructure (agents/, hooks/, lib/, commands/) is write-protected outside the pipeline, and a dispatch sentinel keeps the coordinator from directly editing those paths mid-pipeline — it must re-dispatch the implementer agent (#1296). These are the harness equivalent of guardrails — if the model tries to skip a step, it's physically blocked.
2. Agents (adversarial evaluation) — Each pipeline step is handled by a specialist agent with a specific job and constrained tools. The implementer never reviews its own work — a separate reviewer agent with a skeptical mandate evaluates it (generator/evaluator pattern). Skills power this layer: instead of stuffing every rule into context upfront, domain knowledge is injected progressively — testing standards load during test writing, security patterns during security review — keeping each agent focused within its current step.
3. Continuous Improvement (post-session, self-correcting) — Every session logs JSONL to .claude/logs/activity/. The continuous-improvement-analyst evaluates those logs against PROJECT.md + CLAUDE.md and emits structured finding records; /improve --auto-file promotes recurring findings into GitHub issues labeled auto-improvement. Runs asynchronously, never blocks active work. (This is the layer that, earlier today, caught this very session's gate mis-scopes and filed #1385/#1387/#1388.)
4. Autonomous Self-Improvement (closed-loop, evidence-driven) — /triage clusters the improvement queue by root cause → /drain-queue (and the scheduled cloud-drain) works it through /implement --issues behind safety guardrails → benchmarks gate whether a change is kept or reverted. Effectiveness benchmarks measure reviewer/agent accuracy; HIGH-confidence diagnoses are applied autonomously, the rest filed as issues.

The result: Every feature goes through every step — not because Claude remembers to, but because the harness won't let it skip — and the harness itself gets better every week from its own runtime data.

Recent Improvements

Substantial enforcement hardening shipped recently. Current focus areas:

Mode- and intent-aware enforcement — hooks respect the active pipeline mode (full / light / fix / tdd-first) so a --light run isn't held to the full agent set, and an intent/session-mode classifier scopes the SDLC gates to actual software-development edits (not scratch/exploration/meta-work). The gate also short-circuits message-only git commit --amend — context-aware gating, not surface-signal gating
Default-ON global enforcement — the SDLC gates fire in every repo by default, opt out per-repo with .claude/.bypass (Phase 1 polarity flip, #1142+/#1361)
Autonomous operations loop — /triage clusters the open improvement queue by root cause, /drain-queue drains it through /implement --issues behind safety guardrails, and a scheduled cloud-drain runs the same loop unattended. Cross-machine issue claiming (a GitHub in-progress label + marker-comment mutex) stops a local run and the cloud drainer from racing the same cluster. /goa is a standing infra-health observer
Planning workflow — 7-step /plan + adversarial plan-critic review (1-5 Likert, composite ≥3.0 → PROCEED)
Session analytics — every conversation archived to ~/.claude/archive/ with a 17-column SQLite index
Closed-loop self-improvement — benchmark → modify → re-benchmark → commit-or-revert (see EVALUATION.md)

See CHANGELOG.md for complete release-over-release details with issue references.

The 12 Elements of Harness Engineering

The core insight of harness engineering: reliability in multi-step AI workflows is difficult because failures compound. A 10-step process with 90% accuracy per step fails over 60% of the time. "Agent skills" that are just prompts or markdown files — where you hope the AI follows instructions — lack the dependability required for production work.

The solution is deterministic rails: a software layer that gates and validates every stage. autonomous-dev implements all 12 elements of this framework:

#	Element	How autonomous-dev implements it
1	State Machine	`pipeline_state.py` — 13-phase state machine with `Step` enum, `advance()`/`complete_step()` API, JSON-persisted state
2	Validation Loops	STEP 8 HARD GATE — runs pytest, loops until 0 failures/0 errors. Anti-stubbing gate blocks `raise NotImplementedError()` shortcuts
3	Isolated Sub-Agents	Specialist agents, each spawned with fresh context and constrained tools. Model selection per agent (Haiku/Sonnet/Opus)
4	Virtual File System	Git worktree isolation for batch mode. `checkpoint.py` + `artifacts.py` persist outputs to `.claude/artifacts/` per phase
5	Human-in-the-Loop	Plan mode (STEP 5) requires user approval before implementation. `pause_controller.py` for explicit gates
6	Hook Enforcement	26 hooks with JSON `{"decision": "block"}` hard gates — not prompt-level nudges. Research-confirmed: nudges produce unreliable compliance. Includes `plan_gate.py` (blocks complex Write/Edit without validated plan) and Layer 6 prompt quality gate (blocks anti-pattern prompts to `agents/.md` and `commands/.md`)
7	State Persistence	`CheckpointManager` for resume after failure. `batch_state_manager.py` for multi-feature recovery. `/implement --resume`
8	Context Management	Progressive skill injection loads domain knowledge per-step. `/clear` between features. Agent isolation prevents context rot
9	Deterministic Ordering	`agent_ordering_gate.py` enforces pipeline sequence. "You MUST NOT run STEP 10 before STEP 8 test gate passes"
10	Output Validation	Parallel reviewer + security-auditor in STEP 10. `genai_validate.py` for LLM-as-judge tests. `completion_verifier.py`
11	Observability	`session_activity_logger.py`, `conversation_archiver.py`, `pipeline_timing_analyzer.py`. Structured JSONL + SQLite for long-term analytics
12	Error Recovery	`failure_analyzer.py`, `batch_retry_manager.py`, `stuck_detector.py`, `qa_self_healer.py`. Automatic retry with consent

What's PROJECT.md?

A simple markdown file that defines your project. Example:

# PROJECT.md

## GOALS
- Build a REST API for user management
- Achieve 80% test coverage
- Ship MVP by end of Q1

## SCOPE
IN: User CRUD, authentication, password reset
OUT: Admin dashboard, analytics, billing

## CONSTRAINTS
- Python 3.11+ with FastAPI
- PostgreSQL database
- JWT authentication (no sessions)
- All endpoints must have rate limiting

## ARCHITECTURE
- src/api/ - FastAPI route handlers
- src/models/ - SQLAlchemy models
- src/services/ - Business logic
- tests/ - Pytest test suite

When you run /implement "Add user registration":

Allowed - user registration is IN scope
Claude follows your FastAPI + PostgreSQL + JWT constraints
Tests go in tests/, code follows your architecture

When you run /implement "Add analytics dashboard":

Blocked - analytics is OUT of scope
No code written, no time wasted
You're asked to either change the request or update PROJECT.md

This is how Claude stays aligned - it reads PROJECT.md before every feature.

Install

Prerequisites

Requirement	Install Command
Claude Code 2.0+	Download
Python 3.9+	`python3 --version` to verify
gh CLI (GitHub)	`brew install gh && gh auth login`

One-Line Install

# Per-repo install (default, recommended)
bash <(curl -sSL https://raw.githubusercontent.com/akaszubski/autonomous-dev/master/install.sh)

# Global install (all repos)  
bash <(curl -sSL https://raw.githubusercontent.com/akaszubski/autonomous-dev/master/install.sh) --scope=user

Installation Scopes:

Per-repo (default): Installs to ./.claude/ in current directory. Safer — failures don't affect other repos.
Global (--scope=user): Installs to ~/.claude/ for all repos. Convenient but riskier.

Then fully quit Claude Code (Cmd+Q on Mac, Ctrl+Q on Windows/Linux) and reopen it.

Set Up Your Project

Open your project folder in terminal, then start Claude Code:

cd /path/to/your/project
claude

Run the setup wizard:

/setup

This will walk you through creating PROJECT.md and configuring the plugin for your project.

For existing projects, use:

/align --retrofit

Update

/sync

For consumer Macs enrolled in the auto-update flow: updates are delivered automatically via a launchd timer running scripts/pull-plugin-update.sh. See docs/RUNBOOK.md — Consumer-side auto-update (launchd) for setup.

Uninstall

Two paths — pick the one that matches your situation:

# Normal uninstall (Claude Code still works)
/sync --uninstall

# Shell-only uninstall (Claude Code is broken, hooks are bricked, or /sync itself fails)
bash install.sh --uninstall            # removes hooks and global registration
bash install.sh --uninstall --dry-run  # preview first
bash install.sh --uninstall --repos "/path/to/repo1,/path/to/repo2"  # also clean per-repo settings

See TROUBLESHOOTING.md for the full comparison table and rollback instructions.

Usage

Single Feature

/implement "#72"              # From GitHub issue
/implement "Add user auth"    # From description

Pipeline Modes

/implement "#72"              # Full pipeline (default) - acceptance-first, ~8 agents
/implement --light "#72"      # Light - 4 agents (plan, implement, docs, CIA); no research/security
/implement --tdd-first "#72"  # TDD-first - unit tests before implementation (adds test-master)
/implement --fix "#72"        # Fix - 5-step: root-cause + regression-test + spec-blind HARD GATES

Planning Workflow

Use /plan for changes touching >3 files, >100 lines, or with uncertain approach. The 7-step planning process is enforced by plan_gate.py (blocks complex Write/Edit without a validated plan in .claude/plans/).

/plan "Add OAuth login"       # 7-step structured planning + iterative plan-critic
/plan --no-issues "..."       # Skip auto-creation of GitHub issues
/plan-to-issues               # Thorough-mode fallback for batch issue creation

The /plan flow: problem statement → existing solutions search → minimal path → adversarial critique by plan-critic (1-5 Likert score across 6 axes in /plan; the in-pipeline /implement gate uses a faster 4-axis budget subset, iterating the planner until PROCEED) → on PROCEED, automatically creates GitHub issues when ≥2 independent work items exist. See docs/PLANNING-WORKFLOW.md.

Plan-exit enforcement is default-ON in every repo (since #1361, flipped from #938)

Issue #1361 completed the polarity flip started in the general SDLC enforcement work (#1142+ Phase 1). The plan-exit hooks (plan_mode_exit_detector.py, unified_pre_tool.py) now fire in every repo by default, not just inside autonomous-dev. In prior versions (2026-04-25 through 2026-07-04), foreign projects silently no-op'd — a session could call ExitPlanMode without running plan-critic and nothing objected.

Three plan-review-specific escape hatches exist (any one bypasses this gate only, leaving all other hooks in effect):

/implement --skip-review — one-shot, plan-only. Best for ad-hoc cases.
export AUTONOMOUS_DEV_SKIP_PLAN_REVIEW=1 — persistent across sessions and shells. Recommended for users who want plan-critic permanently disabled.
touch .claude/SKIP_PLAN_REVIEW — local sentinel file. Note: this file is gitignored (covered by .claude/*) and is therefore not committed; it is lost on git clean -fdx or fresh clone. Use the env var for cross-session persistence.

For a whole-hook stack opt-out (turns off plan-exit AND almost every other hook — see the Issue #1435 exception below), use the universal Issue #969 bypass documented below (.claude/.bypass or AUTONOMOUS_DEV_BYPASS=1).

AUTONOMOUS_DEV_GLOBAL_ENFORCEMENT is deprecated (#1361) — enforcement is now the default. Setting the var emits a stderr deprecation notice; it is otherwise a no-op. Remove it to silence the notice.

Universal hook bypass (Issue #969): if any hook is blocking and you cannot run a slash command to unstick it, set AUTONOMOUS_DEV_BYPASS=1 or touch .claude/.bypass — almost every hook honors this and falls through to allow with a logged audit trail. See docs/TROUBLESHOOTING.md.

Exception (Issue #1435 — non-bypassable hard floor): Write/Edit to protected infrastructure inside an autonomous-dev repo (agents/*.md, commands/*.md, hooks/*.py, lib/*.py, skills/*/SKILL.md) is not waived by .claude/.bypass or AUTONOMOUS_DEV_BYPASS=1. _is_protected_infrastructure is a registered hard-floor function (config/hard_floor_hooks.json) and the universal bypass now falls through to the normal protected-infra deny gate instead of short-circuiting it. /implement remains the sole sanctioned path for editing these files; there is no bypass override for this specific gate.

Per-repo enforcement control

autonomous-dev installs hooks into ~/.claude/ globally. Phase 1 (Issue #1142+) flipped the polarity from opt-IN via .claude/.enforce to default-ON subject to a single .claude/.bypass opt-out marker. Three states control per-repo enforcement, in order of precedence:

State	How	Enforcement
Bypass (highest priority)	`touch .claude/.bypass` OR `AUTONOMOUS_DEV_BYPASS=1`	OFF — almost all hooks fall through (Issue #969). Also the supported durable per-repo opt-out. Exception (Issue #1435): protected-infrastructure Write/Edit (`agents/.md`, `commands/.md`, `hooks/.py`, `lib/.py`, `skills/*/SKILL.md`) remains a hard-blocking floor and is NOT waived by this bypass.
Plugin source auto-detect	Repo contains `plugins/autonomous-dev/.claude-plugin/marketplace.json`	ON — autonomous-dev itself (self-maintenance mode relaxes only state-deletion).
Default	None of the above	ON — default-on production-code Write/Edit gate.

Opt a repo out (permanent silence or durable opt-out):

cd /path/to/repo
mkdir -p .claude
touch .claude/.bypass
git add .claude/.bypass
git commit -m "chore: opt out of autonomous-dev SDLC enforcement"

To temporarily re-enable in a bypassed repo: rm .claude/.bypass.

Forgot to remove an emergency bypass? If .claude/.bypass is left in place uncommitted for more than 24h (override via AUTONOMOUS_DEV_BYPASS_STALE_HOURS), every session start prints a non-blocking warning naming the file and its age (Issue #1434). Committing the file (as shown above) is the supported durable opt-out and never warns; the warning exists specifically to catch a forgotten one-off bypass.

The default-on production-code gate (Phase 1, Issue #1142+) blocks Write/Edit to code-file extensions (.py, .ts, .js, .go, .rs, etc.) when no /implement pipeline is active and the edit is non-trivial. The classifier returns one of three tiers:

fix (<20 lines, no AST signal) → directive /implement --fix
light (new function / control-flow / 20-99 lines) → directive /implement --light
full (new class / ≥100 lines) → directive /implement

Test files (test_*.py, files under tests/ or test/) are always exempt. Scratch/ephemeral paths (/tmp/, /private/tmp/, /var/folders/, $SCRATCHPAD, ~/tmp/, ~/.cache/, .claude/tmp/) and any path outside a git worktree or covered by .gitignore are also exempt — pipeline review adds no value for files that are never committed (Issue #1408).

One-shot operator bypass: touch /tmp/skip_write_pipeline_gate (consumed on first check). The sentinel file's contents (or $WRITE_GATE_BYPASS_REASON) are captured as a reason string in the activity log when the bypass is consumed (Issue #1408) — echo "why" > /tmp/skip_write_pipeline_gate records the reason.

Edit/Write to code files remains a hard block. Bash commands that write to code files (cat > X.py, sed -i X.py, tee X.py, heredocs) were downgraded from a hard block to a non-blocking advisory (Issue #1408) — general bash-write detection is unsound (many forms bypass the pattern set), so a hard gate there gave a false sense of security. The command still runs; a [hook advisory] note on stderr recommends the matching /implement variant. Protected-infrastructure Bash writes (agents/*.md, hooks/*.py, lib/*.py, etc.), Edit/Write, and the Issue #803 cross-tool workaround check remain hard gates; git apply and patch < diff are excluded as user-driven patch tooling.

Uninstalling autonomous-dev

Scope	Command	Removes
Project-local	`/sync --uninstall`	`<repo>/.claude/` (hooks, agents, commands, lib, config — anything autonomous-dev deployed)
Globally	`rm -rf ~/.claude/{hooks,lib,config,agents,commands,skills}`	Global plugin install. Warning: this removes ALL Claude Code customisation under those paths, not just autonomous-dev. Back up first.
Per-repo silence (no removal)	`touch .claude/.bypass` (see above)	Nothing — just disables enforcement

/sync --uninstall is the recommended path for clean removal from a single project. The "per-repo silence" approach is best when autonomous-dev is correctly installed but you don't want it active in this particular tree.

Batch Processing

/implement --issues 72 73 74 75    # Multiple GitHub issues with worktree isolation
/implement --issues 72 73 --no-worktree   # In-place (for repos where .claude/* is gitignored)
/implement --batch backlog.txt     # From a file (one feature per line)
/implement --batch backlog.txt --full-tests  # Run the full suite instead of smart test routing
/implement --resume <batch-id>     # Resume after context reset

Project Management

/plan "..."              # 7-step planning workflow with adversarial plan-critic
/status                  # View PROJECT.md goal progress
/align                   # Check alignment (--project, --docs, --retrofit, --content)
/create-issue "..."      # Create GitHub issue with automated research
/plan-to-issues          # Thorough-mode batch issue creation from plan output
/health-check            # Verify plugin installation

Quality & Analysis

/refactor                # Code, docs, and test optimization (--tests, --docs, --docs-redundancy, --code, --fix)
/sweep                   # Quick codebase hygiene (alias for /refactor --quick); --tests for AST-based test pruning analysis
/audit                   # Quality audit (--quick, --security, --docs, --code, --tests)
/skill-eval              # Measure skill effectiveness via behavioral delta scoring (--quick, --skill, --update)
/advise                  # Critical thinking analysis
/improve                 # Automation health analysis (--auto-file to create GitHub issues)
/retrospective           # Session drift detection and alignment updates
/autoresearch            # Autonomous experiment loop — hypothesize, modify, benchmark, commit or revert

Other Commands

/sync                    # Update plugin (--github, --env, --marketplace, --all)
/worktree                # Git worktrees (--list, --status, --merge, --discard)
/scaffold-genai-uat      # Scaffold LLM-as-judge tests into any repo
/mem-search              # Search claude-mem persistent memory
/triage --auto-improvement  # Cluster open auto-improvement issues by root cause; emit ranked work queue (periodic)
/drain-queue             # Autonomous queue drainer wrapping /implement --issues with 6 safety guardrails
/goa                     # Governance, Observability, Audit — autonomous infra-health observer (start | stop | status)

How the Harness Works

Generator / Evaluator Pipeline

Inspired by the adversarial evaluation pattern (generator creates, evaluator judges), every /implement runs this pipeline:

/implement "feature"
     |
PROJECT.md Alignment Check (blocks if misaligned)
     |
Research Self-Critique (inline FEEDBACK pass, Self-Refine pattern)
     |
+------------------+------------------+
| Research-Local   | Research-Web     |  <- Parallel research
| (Haiku)          | (Sonnet)         |
+------------------+------------------+
     |
Planning (Opus)
     |
Plan Validation Gate (plan-critic, Sonnet) <- HARD GATE: adversarial review
     |                                        4 axes (budget mode), PROCEED to pass
     |                                        REVISE re-invokes planner
     |
Acceptance Tests (test-master)            <- Acceptance-first (default)
     |                                       or TDD-first (--tdd-first)
Implementation (Opus)            -> HARD GATE: 0 test failures
     |                           -> HARD GATE: No stubs/placeholders
     |                           -> HARD GATE: Root cause analysis (--fix)
     |                           -> HARD GATE: Hook registration verified
     |
Plan-Implementation Alignment Gate -> HARD GATE: planned vs diff files
     |                               (blocks on >50% divergence)
     |
Spec-Blind Validation (spec-validator) <- HARD GATE: behavioral tests
     |                                    from criteria only, no impl
     |
+----------+------------+-----------+
| Review   | Security   | Docs      |  <- Parallel for low-risk,
| (Sonnet) | (Sonnet)   | (Sonnet)  |     sequential for security-sensitive
+----------+------------+-----------+
     |
Remediation Gate (max 2 cycles on findings)
     |
Continuous Improvement (CIA analysis)    <- dispatched BEFORE commit (#1211)
     |
Agent Completeness Gate (blocks commit if required agents missing)
     |
Git Operations (commit, push, close issue)
     |
Doc-Congruence Gate (HARD GATE: docs/code parity)
     |
Session Archive (transcript + SQLite index)

PROJECT.md Alignment

Features validate against your PROJECT.md before work starts. If you request something OUT of scope, it's blocked - not implemented wrong.

Testing: The Diamond Model

autonomous-dev uses a diamond testing model — not the traditional TDD pyramid. The key insight: with agent-speed code generation, acceptance criteria are the cheapest form of meaningful specification. Unit tests generated by agents are useful as regression locks but poor as specifications — agents game them.

     /  Acceptance Criteria  \     Human-defined, LLM-as-judge evaluated
    / LLM-as-Judge Eval Layer \    Probabilistic, ~80-90% human agreement
   / Integration & Contract    \   Generated from acceptance criteria
   \ Property-Based Invariants /   "Output must be valid JSON", etc.
    \ Deterministic Unit Tests/    Generated by agent, constrained by above
     \  Type System / Lints  /     Hard floor, zero tolerance

What changed from TDD:

Acceptance criteria drive everything — defined before implementation, not unit tests
Unit tests are regression locks, not specifications — they prevent drift, they don't define correctness
LLM-as-judge evaluation validates semantic intent (~85% human agreement, tests/genai/)
Property-based invariants catch bugs unit tests miss (+23-37% pass@1, tests/property/)
Spec-blind validation — a separate agent writes behavioral tests from acceptance criteria without seeing the implementation, then validates against it (STEP 8.5 HARD GATE)

The pipeline default is acceptance-first (/implement "#X"). Legacy TDD-first is available via --tdd-first. See Testing Strategy for the full 6-layer model with research citations.

Context Management

As the context window fills up, models exhibit context anxiety — they rush through steps, declare things done prematurely, and degrade output quality. autonomous-dev handles this with context resets between features:

Each feature uses ~25-35K tokens
After 4-5 features, run /clear to reset context
Run /implement --resume <batch-id> to continue from where you left off

Batch processing handles this automatically with worktree isolation and checkpoint/resume.

Session Analytics & Self-Improvement

autonomous-dev doesn't just run pipelines — it learns from them.

Long-term session analytics: Every conversation is archived to ~/.claude/archive/ with full transcripts and a 17-column SQLite index (per-session tokens, cache hit rates, tool calls, model, per-repo attribution). See docs/SESSION-ANALYTICS.md for schema + queries.

sqlite3 ~/.claude/archive/sessions.db \
  "SELECT project, COUNT(*) sessions,
          SUM(total_output_tokens) out_tok, SUM(cache_read_tokens) cache_read
   FROM sessions GROUP BY project ORDER BY out_tok DESC;"

Closed-loop improvement: The system detects its own weaknesses and fixes them. See docs/EVALUATION.md for the full measurement surface (skill-eval, reviewer benchmark, /improve workflow, autoresearch).

pipeline runs → session logs → /improve detects drift → files GitHub issues
     ↓                                                        ↓
/autoresearch picks up issues → hypothesize → modify → benchmark → deploy if better
     ↑                                                              ↓
     └──────────────────────────────────────────────────────────────┘

/improve analyzes recent sessions for bypasses, test drift, and pipeline degradation — auto-files issues
/autoresearch runs autonomous experiments: modifies agent prompts, benchmarks the change, commits if it improves quality or reverts if it doesn't
/retrospective detects intent evolution and proposes alignment updates
/skill-eval measures whether skill injections actually change agent behavior (behavioral delta scoring)

The result: The harness gets better every week without anyone thinking about it.

What You Get

Component	Count	Purpose
Commands	23	User-facing slash commands for workflows
Agents	16	Specialized AI for each SDLC stage, model-tiered (Haiku/Sonnet/Opus)
Skills	20	Domain knowledge injected progressively per-step
Hooks	26	Automatic validation and enforcement (JSON `{"decision": "block"}` hard gates)
Libraries	241	Python utilities

Documentation

Core Concepts

Architecture - Four-layer system (hooks + agents + continuous improvement + autonomous self-improvement)
Harness Evolution - How the harness has evolved across releases
Maintaining Philosophy - Why alignment-first works
Model Behavior Notes - Production-validated patterns for prompt design

Workflows

Planning Workflow - 7-step /plan + plan-critic adversarial review
Pipeline Modes - /implement mode matrix (full / light / fix / batch) with per-mode agent sets
Batch Processing - Multi-feature workflows with worktree isolation
Git Automation - Auto-commit, push, issue close
Workflow Discipline - Pipeline ordering enforcement (two-tier)

Observability & Self-Improvement

Session Analytics - sessions.db schema, archive layout, query recipes
Evaluation - Skill-eval, reviewer benchmark, the closed-loop self-improvement cycle

Prompt & Agent Design

Prompt Engineering - Constraint budgets (MOSAIC), register shifting, HARD GATE patterns
Agents Reference - All 16 specialist agents with model tiers
Skills Reference - All 20 progressive skill packages with trigger conditions

Testing

Testing Strategy - Diamond testing model (acceptance-first, 6 layers)
GenAI Tests - LLM-as-judge semantic validation
Property Tests - Hypothesis-based invariants
Spec Validation - Spec-blind behavioral tests

Reference

Commands - All 23 user-facing commands
Hooks - 26 active hooks
Hook Registry - Sidecar metadata schema
Libraries - 241 Python utilities
Scripts - Operational tooling (deploy, validate, benchmark, mine)
Label Taxonomy - GitHub labels + [BYPASS]/[INCOMPLETE]/[ORDERING] finding tags

Security

Sandboxing - Command classification and shell injection detection
MCP Architecture - Per-repo .mcp.json default, token-bleed prevention, migration helper
MCP Security - Path traversal, command injection, SSRF prevention
Security Audit - Security scanning architecture

Troubleshooting

Troubleshooting Guide - Common issues

Quick Reference

# Install
bash <(curl -sSL https://raw.githubusercontent.com/akaszubski/autonomous-dev/master/install.sh)
# Restart Claude Code (Cmd+Q / Ctrl+Q)

# Daily workflow
/implement "#72"                    # Single feature
/implement --issues 1 2 3           # Multiple features
/implement --fix "#99"              # Fix a failing test
/clear                              # Reset context between features

# Check status
/health-check                       # Verify installation
/status                             # View alignment

# Update
/sync                               # Get latest version

Support

Issues: github.com/akaszubski/autonomous-dev/issues
Changelog: CHANGELOG.md

License

MIT License

Name		Name	Last commit message	Last commit date
Latest commit History 1,983 Commits
.claude		.claude
.github		.github
.hooks		.hooks
.mcp		.mcp
baselines		baselines
custom		custom
docs		docs
logs		logs
plugins/autonomous-dev		plugins/autonomous-dev
scripts		scripts
tests		tests
.env.example		.env.example
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
CONTRIBUTING.md		CONTRIBUTING.md
PROJECT.md		PROJECT.md
README.md		README.md
VERSION		VERSION
install.sh		install.sh
pytest.ini		pytest.ini
setup.cfg		setup.cfg

Folders and files

Latest commit

History

Repository files navigation