Agent_Do is a public, agent-agnostic skill pack for strict, spec-driven, multi-agent software workflows. It works with Codex, Claude Code, and any other tool that can read skills.
It gives you visible /agent-do ... launcher commands plus a file-backed coordination system for planning, assigning, handing off, reviewing, repairing, and archiving work across multiple AI coding tools.
- One command family: every workflow command starts with
/agent-do. - Product-agnostic roles: Codex, Claude Code, Antigravity, OpenCode, Cursor, Cline, or any capable agent can plan, execute, review, or repair.
- Strict gates: plan compliance, claims, code review, security review, design review, rule checks, validation evidence, and token-budget checks.
- File-backed handoff:
AI_COORDINATION/is the source of truth instead of hidden chat history. - Cross-platform helpers: Python scripts work on Windows, macOS, and Linux; PowerShell helpers remain available on Windows.
| Path | Purpose |
|---|---|
skills/agent-workflow-coordinator |
Main skill with protocols, templates, scripts, and review gates. |
skills/agent-do-* |
Small launcher skills that expose /agent-do ... entries in Codex. |
docs/ |
User documentation and command references. |
tests/ |
Unit tests (stdlib unittest) for the helper scripts and linter. |
tools/ |
Maintainer helpers not part of the published skill (e.g. sync_skill.ps1). |
scripts/validate_repository.py |
Repository-level validation used by CI. |
.github/workflows/validate.yml |
GitHub Actions validation for Python and PowerShell paths. |
Copy the folders under skills/ into your agent's skills directory, then restart the agent.
Codex — ~/.codex/skills (Windows: %USERPROFILE%\.codex\skills):
mkdir -p "$HOME/.codex/skills" && cp -R skills/* "$HOME/.codex/skills/"Copy-Item -Recurse .\skills\* "$env:USERPROFILE\.codex\skills\"Claude Code — ~/.claude/skills (the agent-workflow-coordinator skill is the one Claude uses; the agent-do-* launchers are a Codex UI convenience):
mkdir -p "$HOME/.claude/skills" && cp -R skills/agent-workflow-coordinator "$HOME/.claude/skills/"Copy-Item -Recurse .\skills\agent-workflow-coordinator "$env:USERPROFILE\.claude\skills\"Other agents — copy skills/agent-workflow-coordinator into whatever directory your agent loads skills from; the workflow is driven by SKILL.md and the Markdown protocols, not by any product-specific feature.
On Windows, tools/sync_skill.ps1 can mirror the skills into one or more directories at once (use -IncludeClaude to also target Claude Code).
Restart the agent after installing or updating skills.
Type /agent-do in Codex and choose one of the launcher entries:
| Command | Use |
|---|---|
/agent-do init |
Create project coordination files. |
/agent-do status |
Summarize active tasks, owners, blockers, and next action. |
/agent-do constitution |
Create or update strict project rules. |
/agent-do spec |
Write a user-focused feature specification. |
/agent-do clarify |
Resolve high-impact ambiguity. |
/agent-do checklist |
Generate requirements-quality checks. |
/agent-do plan |
Create a decision-complete implementation plan. |
/agent-do tasks |
Generate or refresh task breakdown. |
/agent-do analyze-artifacts |
Check consistency across spec, plan, and tasks. |
/agent-do split |
Split work across non-overlapping agents. |
/agent-do assign |
Assign task ownership and scope. |
/agent-do handoff |
Prepare a compact executor prompt. |
/agent-do review |
Run general workflow review. |
/agent-do code-review |
Review correctness and maintainability. |
/agent-do security-review |
Review security-sensitive changes. |
/agent-do design-review |
Review UI/UX and visual quality. |
/agent-do rules-check |
Enforce user and project rules. |
/agent-do repair |
Write a focused repair prompt. |
/agent-do compact |
Reduce context usage and refresh context packs. |
/agent-do archive |
Archive completed task artifacts. |
/agent-do init
/agent-do constitution strict project rules
/agent-do spec describe the requested feature
/agent-do clarify
/agent-do plan TASK-001
/agent-do tasks
/agent-do assign TASK-001 Antigravity "src/ui/**"
/agent-do handoff TASK-001 Antigravity
/agent-do review TASK-001
Roles are assigned by capability and task scope, not by product name.
| Role | Allowed agents |
|---|---|
| Planner | Any capable agent assigned to planning. |
| Coordinator | Any capable agent assigned to coordination files. |
| Executor | Any capable agent assigned to the write scope. |
| Reviewer | Any capable independent agent that did not execute the same task. |
| Security Reviewer | Any capable independent security reviewer. |
| Design Reviewer | Any capable independent UI/UX reviewer. |
| Repair Executor | Any capable agent assigned to the repair scope. |
The executor for a task must not approve its own work.
The repository includes GitHub Actions validation, but if Actions are disabled or blocked on the GitHub account, use the local validation command below.
Run the full repository validation:
python scripts/validate_repository.pyRun the main skill lint directly:
python skills/agent-workflow-coordinator/scripts/lint_skill.pyWindows PowerShell lint:
powershell -ExecutionPolicy Bypass -File skills\agent-workflow-coordinator\scripts\lint_skill.ps1This project follows semantic versioning. Changes are recorded in CHANGELOG.md and the skill-level changelog. Tagged releases are published under Releases; pin a tag (for example v1.3.0) for a stable install.
MIT. See LICENSE.