autonomous-dev

Dedicated research/audit workflow for security audits, code analysis, and exploration tasks

Cancel autonomous execution and clean up

Resume autonomous execution from checkpoint after session interrupt or token limit

Execute a prepared autonomous plan. Supports --overnight for fully autonomous execution, --continue for resuming. Reads state from .claude/auto-execution/ created by /auto-prepare.

Check autonomous-dev system health and diagnose issues

Use this agent to review code changes after autonomous implementation. This agent analyzes completed tasks for quality, security, and adherence to project conventions. Examples: <example> Context: A task has been completed and needs review before merging user: "Review the changes in auto/task-1 branch" assistant: "I'll use the code-reviewer agent to analyze the changes for quality and issues." <commentary> After task completion, code-reviewer should be used to catch issues before integration. </commentary> </example> <example> Context: All parallel tasks completed, ready for final review assistant: "Before merging, I'll run the code-reviewer agent to ensure quality across all branches." <commentary> Code review is especially important before integrating multiple parallel changes. </commentary> </example>

Use this agent BEFORE merging parallel task branches. This agent creates a temporary merge, runs full verification, and validates schema/contracts before allowing the actual merge. Examples: <example> Context: Multiple parallel tasks completed in worktrees assistant: "All tasks complete. Running integration-validator before merge." <commentary> Integration validation is mandatory before Phase 5 (Integrate). </commentary> </example> <example> Context: One task changed database schema, another uses the table assistant: "Schema changed in task-1. Using integration-validator to verify task-2 compatibility." <commentary> Integration validator catches cross-branch incompatibilities. </commentary> </example> <example> Context: Parallel tasks may have conflicting changes user: "Verify all branches can merge safely" assistant: "I'll use integration-validator to test the combined changes." <commentary> Pre-merge validation prevents post-merge surprises. </commentary> </example>

Independent QA agent for verifying task completion. This agent runs in a fresh environment to validate that a task truly works, independent of the task-executor. Use this agent after task-executor signals READY_FOR_QA. The QA agent provides unbiased verification by testing in isolation. <example> Context: A task-executor has completed a feature and signals ready for QA user: "task-executor signals READY_FOR_QA: task-3 (UserForm component)" assistant: "I'll dispatch the qa-agent to independently verify this task." <commentary> The qa-agent verifies the task in a fresh environment, preventing confirmation bias from the task-executor's own testing. </commentary> </example> <example> Context: Multiple tasks completed, need QA before integration user: "All parallel tasks signaled READY_FOR_QA" assistant: "Dispatching qa-agent for each completed task to verify independently." <commentary> Each task gets independent QA verification before being approved for merge. </commentary> </example>

Use this agent when code touches database queries, types, or schema. This agent validates that TypeScript/code types match the actual database schema using MCP tools (primarily Supabase). Examples: <example> Context: A developer added a new database column in Supabase user: "I added a 'role' column to the users table, check if my types are correct" assistant: "I'll use the schema-validator agent to verify your TypeScript types match the database." <commentary> Use schema-validator when database schema changes to catch type mismatches early. </commentary> </example> <example> Context: Writing code that queries the database assistant: "Before committing, let me validate the schema with schema-validator agent." <commentary> Proactively use schema-validator after writing database-related code. </commentary> </example> <example> Context: Integration validation phase assistant: "Running schema-validator as part of integration checks." <commentary> Schema validation is part of the integration validation pipeline. </commentary> </example>

Use this agent to execute a single task from an autonomous development plan. This agent works in an isolated git worktree and implements one atomic task with full verification. Examples: <example> Context: Autonomous execution has a task ready to be implemented user: "Execute task 3: Create UserForm component" assistant: "I'll use the task-executor agent to implement this task in an isolated worktree." <commentary> The task-executor agent should be used when there's a specific, well-defined task from a decomposed plan that needs implementation. </commentary> </example> <example> Context: Multiple parallel tasks need execution assistant: "I'm dispatching task-executor agents for the parallel group: task-1, task-2, task-3" <commentary> Multiple task-executor agents can run in parallel, each in its own worktree. </commentary> </example>

This skill should be used after code implementation to assess semantic code quality. Goes beyond linting to analyze maintainability, complexity, architecture compliance, and overall code health using LLM-based analysis. Invoked automatically in Phase 6 or when user asks to "review code quality", "analyze maintainability", "check complexity".

This skill should be used when the user asks to "resolve merge conflicts", "fix git conflicts", "merge branches", or when autonomous-dev encounters merge conflicts during branch integration. Intelligently resolves git conflicts or escalates to user when uncertain.

This skill should be used when the work type is FRONTEND and visual verification is needed. Provides browser automation for screenshot comparison, responsive testing, and interactive element verification using Playwright or Puppeteer.

This skill should be invoked at the START of any /auto-execute session. It reads state files from .claude/auto-execution/ and establishes context for continuing autonomous execution. Provides fast context bootstrap with minimal token usage (~1KB vs ~50KB of full context).

This skill defines the Initializer Agent protocol for the Two-Agent Pattern. The Initializer Agent runs ONCE at the start of a new task to analyze, plan, and set up state for Coding Agents. Use this when starting a new autonomous development task that requires session boundaries.

10 hooks across 4 events

Structured thinking for complex problem solving