Skip to content

docs

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

docs

README.md

Documentation

AgentSys is a modular runtime and orchestration system for AI agents. These docs cover the architecture, commands, and workflows — how agents compose into pipelines, how phases gate execution, and how state persists across sessions.

New here? Start with USAGE.md to see commands in action.

Quick Links

I want to... Go to
Install and start using INSTALLATION.md
See examples and workflows USAGE.md
Understand how /next-task works workflows/NEXT-TASK.md
Understand how /ship works workflows/SHIP.md
Run /perf investigations perf-requirements.md
Use with OpenCode or Codex CROSS_PLATFORM.md
See all slop patterns reference/SLOP-PATTERNS.md
See all agents reference/AGENTS.md
Understand the architecture ARCHITECTURE.md

Document Categories

Getting Started

Document Description
INSTALLATION.md Install via marketplace or npm. Prerequisites. Verification.
USAGE.md Command examples, common workflows, tips.

Workflow Deep-Dives

Document Description
workflows/NEXT-TASK.md Complete /next-task flow: phases, agents, state management, resume.
workflows/SHIP.md Complete /ship flow: CI monitoring, review handling, merge, deploy.
perf-requirements.md /perf rules and required phases.
perf-research-methodology.md /perf process details, benchmarking method.

Reference

Document Description
reference/AGENTS.md All 42 agents: purpose, model, tools, restrictions.
reference/SLOP-PATTERNS.md All detection patterns by language, severity, auto-fix.

Platform & Architecture

Document Description
CROSS_PLATFORM.md Claude Code, OpenCode, Codex CLI setup. Migration.
ARCHITECTURE.md Directory structure, libraries, state management.

Key Concepts

Commands

Command Purpose
/next-task Task discovery → implementation → review → ship
/ship Push → PR → CI → reviews → merge → deploy
/deslop 3-phase slop detection and cleanup
/audit-project Multi-agent code review
/drift-detect Compare docs to actual code
/repo-map Build cached AST repo map
/perf Performance investigation workflow
/enhance Analyze prompts, plugins, agents, docs, hooks, skills
/sync-docs Sync docs with code changes

Internal Skills

  • orchestrate-review — Defines review passes and signal thresholds for the Phase 9 review loop

State Files

File Location Purpose
tasks.json {state-dir}/ Which task is active
flow.json {state-dir}/ (worktree) Which phase you're in
preference.json {state-dir}/sources/ Cached task source preference
repo-map.json {state-dir}/ Cached AST repo map

State directories by platform:

  • Claude Code: .claude/
  • OpenCode: .opencode/
  • Codex CLI: .codex/

Agent Models

Model Used For
opus Complex reasoning (exploration, planning, implementation, review)
sonnet Pattern matching (slop detection, validation, discovery)
haiku Mechanical execution (worktree, simple-fixer, ci-monitor)

Certainty Levels

Level Meaning Auto-Fix?
CRITICAL Security issue Yes (with warning)
HIGH Definite problem Yes
MEDIUM Probable problem No
LOW Possible problem No

Diagnostics Scripts

When working in the repo directly, you can sanity-check detection and tooling:

npm run detect   # Platform detection (CI, deploy, project type)
npm run verify   # Tool availability + versions

Getting Help