A centralized, reusable library of high-leverage skills, specialized subagents, and automated developer workflows for AI pair programming assistants (Google Antigravity, Gemini CLI, Claude Code, OpenAI Codex, Cursor, GitHub Copilot, and Roo Code / Cline).
Instead of recreating session scripts, agent definitions, and context rules across separate repositories, agent-toolkit provides a single source of truth that maintains project continuity, enforces engineering discipline, minimizes token consumption, and eliminates repetitive context synchronization.
Install agent-toolkit globally and configure the unified atk CLI:
macOS / Linux (Bash / Zsh):
curl -fsSL https://raw.githubusercontent.com/raddadengineer/agent-toolkit/main/install.sh | bashWindows / Cross-Platform (PowerShell):
irm https://raw.githubusercontent.com/raddadengineer/agent-toolkit/main/install.ps1 | iexConfigure Any Project in 5 Seconds:
cd ~/my-project
atk install # Symlinks configs for all AI tools into your project
atk start # Prime session: branch check, context sync & memory recallπ For complete setup options, manual cloning, and team copy modes, see the Installation & Setup Guide.
| Guide | Description |
|---|---|
| π§° Detailed Component Reference | Deep dive into all 18 skills, 11 subagents, lifecycle hooks, and MCP servers. |
| β‘ CLI Commands & Terminal Recipes | Standalone terminal recipes for atk, token compression, memory, PR auditing, and debugging. |
| π οΈ Installation & Setup Guide | Assistant-specific recipes, symlink vs copy modes, global PATH configuration, and multi-project updating. |
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β 1. SESSION KICKOFF β
β atk start (or /session-starter) β
β β’ Ingests previous handover from SESSION_LOG.md β
β β’ Verifies git branch health and working tree state β
β β’ Validates context file sync (AGENTS.md, CLAUDE.md, GEMINI.md) β
β β’ Recalls active project patterns and anti-patterns β
ββββββββββββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββββββββββ
β
ββββββββββββββββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββ
β 2. CONTEXT OPTIMIZATION (85β95% Token Savings) β
β atk profile / atk pack (or /compress-context) β
β β’ Profiles token consumption across directories β
β β’ Extracts lightweight AST interface skeletons β
β β’ Prevents context rot and stays within LLM attention limits β
ββββββββββββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββββββββββ
β
ββββββββββββββββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββ
β 3. SURGICAL CODING, TESTING & ADVERSARIAL REVIEW β
β β’ Software Engineer orchestrates surgical modifications β
β β’ Test Engineer scaffolds deterministic Red-Green test proofs β
β β’ Code Reviewer audits diff for error swallowing, concurrency & security β
β β’ Memory Manager recalls established rules (atk memory recall) β
ββββββββββββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββββββββββββ
β
ββββββββββββββββββββββββββββββββββββββββΌβββββββββββββββββββββββββββββββββββββββ
β 4. PRE-FLIGHT PR AUDIT & DISCIPLINED CLOSURE β
β atk pr / atk close (or /session-closer) β
β β’ Audits diff for debug logs and leaked credentials β
β β’ Enforces Zero False Verification (test exit code 0) β
β β’ Prepends structured handover log to SESSION_LOG.md β
β β’ Out-of-band Dreamer distills session learnings into memory store β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
All skills are delivered via progressive disclosure (.agents/skills/), and specialized subagents operate in isolated contexts (.agents/agents/):
| Skill | Primary Focus | CLI Shortcut | Reference |
|---|---|---|---|
session-starter |
Session kickoff, handover ingestion & memory priming | atk start |
Docs |
session-closer |
Handover ledger logging, context sync & verification | atk close |
Docs |
context-optimizer |
Token heatmap profiling & AST interface skeletonizing | atk profile / atk pack |
Docs |
memory-manager |
Persistent scoped developer memory & SHA-256 recall | atk memory |
Docs |
dreamer |
Out-of-band offline transcript analysis & distillation | atk dream |
Docs |
pr-prep |
Pre-flight secret/log sanitization & PR descriptions | atk pr |
Docs |
code-reviewer |
5-dimension adversarial code audit & verdicts | atk review |
Docs |
test-engineer |
Deterministic test scaffolding & boundary matrices | atk test-gen |
Docs |
system-architect |
Architecture Decision Records (ADRs) & boundaries | atk adr |
Docs |
debugger |
Stack trace deobfuscation, repro.sh & git bisect | atk debug |
Docs |
security-auditor |
SAST vulnerability scanning & secret leak detection | atk audit |
Docs |
database-engineer |
Reversible migrations & table lock contention checks | atk db |
Docs |
container-engineer |
Hardened multi-stage Dockerfiles & caching linters | atk docker |
Docs |
ci-workflow |
GitHub Actions pipeline synthesis & action unpinning | atk ci |
Docs |
dependency-auditor |
Multi-ecosystem wildcard & deprecation auditing | atk dep |
Docs |
release-manager |
SemVer bump calculation & Conventional Commits changelogs | atk release |
Docs |
doc-sync |
Documentation, schema parity & broken link sentinel | atk drift |
Docs |
sys-admin |
Host health triage, server/IaC hardening & SRE templates | atk sysadmin |
Docs |
| Subagent | Persona & Focus | Reference |
|---|---|---|
software_engineer |
Tech Lead orchestrator, surgical changes, peer-review gating, Zero False Verification. | Docs |
code_reviewer |
Read-only adversarial diff auditor (security, concurrency, error swallowing, churn). | Docs |
test_engineer |
Test automation specialist, deterministic fixtures, mocks, and Red-Green proof. | Docs |
system_architect |
Interface contract design, modular boundary enforcement, and ADR synthesis. | Docs |
context_optimizer |
Token consumption profiling, AST skeleton generation, and hybrid packing. | Docs |
debugger |
Root-cause analyzer, minimal reproduction scripts (repro.sh), automated git bisect. |
Docs |
security_auditor |
Application security scanner, injection checks, and credential leak inspection. | Docs |
database_engineer |
Zero-downtime DDL, lock contention validation, and migration reversibility. | Docs |
devops_engineer |
Hardened Dockerfiles, GitHub Actions synthesis, and SemVer release management. | Docs |
research |
Isolated documentation reading, RFC exploration, and version-pinned snippet distillation. | Docs |
sys_admin |
USE/RED triage, log condensation, safe rollback apply, host drift & IaC hardening. | Docs |
agent-toolkit compiles your skills, instructions, and rules into the native configuration format of every leading AI coding assistant:
| Tool | Export Path | Native Integration |
|---|---|---|
| Claude Code | .claude/commands/CLAUDE.md |
18 interactive slash commands (/session-starter, /code-review, /memory-manager, /sys-admin, etc.) |
| OpenAI Codex | .codex/config.toml.codex/agents/*.toml.agents/skills/, AGENTS.md |
18 progressive skills, 11 sandboxed TOML subagents (read-only / workspace-write) & MCP config |
| Cursor | .cursor/rules/*.mdc.cursorrules |
18 Cursor 2.0 MDC rules with glob trigger conditions and engineering standards |
| GitHub Copilot | .github/copilot-instructions.md.github/prompts/ |
Global repo instructions and 18 reusable .prompt.md slash templates |
| Roo Code / Cline | .roomodes.clinerules |
11 multi-agent JSON personas (software-engineer, code-reviewer, devops-engineer, sys-admin, etc.) |
| Windsurf | .windsurfrules |
Cascade rules for surgical modifications and session continuity |
| Antigravity / Gemini CLI | .agents/skills/.agents/agents/ |
Native 18-skill progressive disclosure and 11 specialized subagents |
π For assistant-specific recipes and setup commands, see the Installation Guide.
agent-toolkit/
βββ bin/
β βββ atk # Unified global CLI launcher (Bash/Zsh)
β βββ atk.ps1 # Unified global CLI launcher (PowerShell)
βββ install.sh / install.ps1 # Multi-tool installers and symlink managers
βββ skills/ # 18 progressive disclosure skills
βββ agents/ # 11 specialized subagent configurations
βββ memory/ # Persistent repository memory (patterns, rules, ADRs)
βββ docs/ # Detailed component, CLI, and installation guides
β βββ components.md # Comprehensive reference for skills & subagents
β βββ cli.md # Standalone recipes & terminal commands
β βββ installation.md # Multi-tool installation & registry guide
βββ scripts/ # Central CLI engines, adapters & sentinels
βββ hooks/ # Antigravity PreToolUse and PostToolUse hooks
βββ mcp/ # Native MCP suite: Playwright, GitHub, SQLite
βββ tests/ # 158 deterministic unit tests (Zero False Verification)
Verify the entire toolkit's synchronizers, auditors, adapters, and memory engines:
python3 -m unittest discover -s testsMIT License. Free to use across personal and commercial projects.