agent-dev-toolkit
Health Warn
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 6 GitHub stars
Code Fail
- exec() — Shell command execution in adapters/opencode/assets/plugins/agent-dev-toolkit-marker.js
Permissions Pass
- Permissions — No dangerous permissions requested
No AI report is available for this listing yet.
Multi-agent skills toolkit: sync shared SDD workflows, stack skills, and policy into Cursor, Claude, Codex, Copilot, and more.
# Agent Dev Toolkit
Unified multi-agent skills core with per-agent adapters. Sync the same SDD workflows, stack skills, and policy into Cursor, Claude Code, Codex, GitHub Copilot, Antigravity, OpenCode, Grok Build, ZCode ADE, Hermes, and OpenHands.
Public — clone and fork freely; no upstream contributions; GitHub Issues are bugs only (see CONTRIBUTING.md and docs/REPO_GOVERNANCE.md). Security: SECURITY.md.
Prerequisites
Supported OS: Windows, Linux (Ubuntu, Debian, and derivatives), and macOS.
| Requirement | Notes |
|---|---|
| PowerShell | Windows: PowerShell 5.1+ or pwsh 7+ (recommended). Linux / macOS: pwsh 7+ only (Windows PowerShell 5.1 is not available on those OS). Install PowerShell |
| Git | Clone / update this repo |
| Target agent | At least one supported agent (see table below) |
Full install steps: docs/INSTALL.md.
What this is
| Layer | Role |
|---|---|
| Core | Agent Skills (core/skills/), policy, router, SDD contracts — agent-neutral |
| Adapters | Publish core into each agent’s install layout (~/.cursor, ~/.claude, …) |
| CLI | toolkit.ps1, sync-agent.ps1, validate-agent.ps1 |
| Validation | In-repo fixtures + CI smokes — no live home write required for green |
Quick start
git clone https://github.com/tibursocampos/agent-dev-toolkit.git agent-dev-toolkit
cd agent-dev-toolkit
# Primary entry — interactive Smart Manager
pwsh -NoProfile -File .\scripts\toolkit.ps1
In the menu:
- Sync agent — pick an agent, then live agent home (Enter = default; writes under your profile after confirm), in-repo fixture (safe), or custom path.
- Validate core only —
validate-corerepo contracts with no home write. - Help and docs — what each action does and equivalent
-Actionflags.
Advanced / scripting flags (-Action, -Agent, -InstallRoot, -AllowUserHome, -Mode) and CI smokes (Invoke-CursorCiSmoke, Invoke-ClaudeCiSmoke, …): docs/INSTALL.md, docs/VALIDATION.md.
First use end-to-end: docs/guides/01-getting-started.md.
Use skills (after sync)
Open a consumer project in your agent and invoke skills by id (slash / when the host supports it):
| Workflow | Invoke |
|---|---|
| Catalog (all adapters) | help-skills → static CATALOG.md + OPERATOR.md |
| Classic SDD | sdd-spec → sdd-plan → sdd-develop |
| Backlog Refine | refine-story → split-story-checklist |
| Stack shortcut | developer or dotnet-developer / react-developer / … |
| Orchestrated Delivery | memory-bank-init → orchestrate-analyze → orchestrate-deliver → orchestrate-develop |
| Greenfield ARCH | Via Orchestrated Delivery analyze: architect draft → sim confirm (not a slash skill) |
Same skill call flow; internal contracts (REQ, validate, CHANGE, EVD, STATE, TRACE, selective retrieval, navigation ## Related, clarification readiness B/I, develop session gate) are gates/artifacts inside those skills — not new slash skills or a second toolkit. SQLite/FTS is out of scope as a deliverable.
Parallel specialists for multi-facet work are the router default after sync (see core/router/AGENTS.md). Chat and persisted artifacts follow the user chat language; spawn prompts stay en-US (core/skills/_shared/agents/LANGUAGE.md). Catalog: docs/SKILLS.md. Daily decision tree: docs/guides/README.md. Credits: docs/CREDITS.md.
Supported agents
| id | Display name | Typical live root | Notes |
|---|---|---|---|
cursor |
Cursor | ~/.cursor |
Skills, .mdc rules, hooks; path/secrets preToolUse + beforeShellExecution (failClosed, GuardCommon); agents/ |
antigravity |
Antigravity | ~/.gemini |
Official config/*; PreToolUse path/secrets under config/hooks |
claude |
Claude Code | ~/.claude |
Skills, rules .md, CLAUDE.md, settings merge; PreToolUse path/secrets deny |
codex |
Codex | ~/.codex |
Dual-root plugin + rules; PreToolUse path/secrets; agents/*.toml; optional -UserScope; default plugin-only |
copilot |
GitHub Copilot | ~/.copilot or .github |
-Mode user|repo; hooks version:1 preToolUse path/secrets |
opencode |
OpenCode | ~/.config/opencode |
Hooks = JS tool.execute.before path/secrets throw; agents=true → InstallRoot/agents/ |
grok |
Grok Build | ~/.grok |
Native publish; PreToolUse path/secrets; agents=true → InstallRoot/agents/ |
zcode |
ZCode (Z.ai ADE) | ~/.zcode |
ADE filesystem; PreToolUse path/secrets; not GLM Coding Plan |
hermes |
Hermes | $HERMES_HOME |
Plugin + shell agent-hooks path/secrets; keyed config.yaml only; never SOUL/tokens/gateway |
openhands |
OpenHands | Project tree; user ~/.agents |
Shell pre_tool_use + guard_pre_tool.sh (fail-closed); subagents=none |
Per-agent contract: docs/ADAPTERS.md. Shared guard: adapters/_shared/guard-rules.md + GuardCommon.ps1.
Skills preview (40)
| Group | Examples |
|---|---|
| Classic SDD | sdd-spec, sdd-plan, sdd-develop, read-sdd-artifact |
| Backlog Refine | refine-story, split-story-checklist |
| Orchestrated Delivery | memory-bank-init, orchestrate-analyze, orchestrate-deliver, orchestrate-develop |
| Stack | developer, dotnet-developer, java-developer, react-developer, angular-developer, vue-developer, … |
| Docs RAG | document-plan, document-implement |
| Ops | help-skills, code-review, commit, push, open-github-pr, test-coverage, repair-dotnet-build, … |
Full list: docs/SKILLS.md · agent SoT: help-skills → _shared/skills-catalog/CATALOG.md + OPERATOR.md.
Documentation
| Doc | Content |
|---|---|
| docs/README.md | Documentation index / TOC |
| docs/overview.md | Architecture overview (RAG-friendly) |
| docs/INSTALL.md | Clone, sync, live home, uninstall |
| docs/VALIDATION.md | validate-core + keyed uninstall asserts + AllowUserHome forward + 10 agent smokes |
| docs/SKILLS.md | Skill catalog |
| docs/CREDITS.md | Third-party inspiration (Caveman, Impeccable, Spec Kit) |
| docs/guides/07-caveman-mode.md | Caveman default OFF, commands, levels |
| docs/guides/08-orchestrator-mode.md | Orchestrator charter, orchestrator_mode |
| docs/guides/README.md | Decision tree + guides |
| docs/ARCHITECTURE.md | Layers and install layouts |
| docs/ADAPTERS.md | Adapter registry and per-agent publish |
| CONTRIBUTING.md | Clone/fork OK; no upstream PRs; Issues = bugs only |
| docs/REPO_GOVERNANCE.md | OSS audiences + policy pointers |
| SECURITY.md | Vulnerability reporting (no invented private email) |
Domains
| Domain | Content |
|---|---|
| docs/domains/core.md | Skills, policy, router, SDD |
| docs/domains/adapters.md | Registry + adapters |
| docs/domains/cli-scripts.md | toolkit / sync / validate |
| docs/domains/validation-ci.md | Fixtures and CI workflow |
Repository layout
agent-dev-toolkit/
core/ # skills, policy, router, SDD contracts
adapters/ # registry.json + per-agent modules
scripts/ # toolkit.ps1, sync-agent, validate-agent, validation/
docs/ # public documentation
memory-bank/ # durable workspace map (Orchestrated Delivery Step 0)
.github/workflows/ # validate-toolkit.yml (+ enforce-release-source.yml)
README.md
CONTRIBUTING.md
SECURITY.md
Safety
- Default
InstallRootis an in-repo fixture — CI never writes under%USERPROFILE%. - Live agent home requires
-AllowUserHome. - Uninstall is keyed (toolkit artifacts only) for all adapters — not a wholesale home wipe. Preserves
sdd/sessionsandsdd/manifest.json.
CI runs validate-core, keyed uninstall asserts, Assert-SyncAllowUserHomeForward, plus all ten agent CI smokes on push/PR to develop / master / main (see .github/workflows/validate-toolkit.yml). Release PRs into master/main must come from develop (enforce-release-source.yml).
License
MIT © 2026 Raphael Campos. See CONTRIBUTING.md for contribution policy (clone/fork OK; no community PRs).
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found