Project Setup: Spec, Risk-First Backlog & Verifier
Before your coding agent writes a line of product code, get a classified plan, a risk-ordered backlog with checkable Done-when criteria, unified agent instructions, and a verifier that fails loudly.
Free
Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+21 moreClaude CodeCursorCodex CLIMuseDots
Manus
Grok
Grok Bot+18 agentsProject Setup: Spec, Risk-First Backlog & Verifier
Example session with this skill installed
Set up this repo for the story flow using the project-setup skill. It is an existing Python library. Story ID prefix: CAN.
- Read your context and instructions
- Compiled the project setup
Summary
The repo scan is done and nothing has been written yet. Existing-project mode needs your approval before any file changes, so here are the findings and the plan.
Real run from our install test on a small Python library, lightly trimmed.
Proposed classification
| Mode | existing |
| Product type | SDK/library (Python, requires-python >=3.9) |
| Lifecycle | prototype |
| Surfaces | public Python API only: no UI, CLI, CI or release pipeline |
| Story ID prefix | CAN (CAN-001, CAN-002) |
| Delivery | dual-agent: Claude implements, Codex reviews, one fix pass, you open the PR and merge |
Gap analysis
| Area | Class | Evidence | Proposed action |
|---|---|---|---|
| Source and tests | Adopt | textstats/ plus a 2-case unittest suite | none |
| Verifier | Adapt | the scaffold drafted pytest -q, which isn't installed | scripts/verify-project.sh runs python3 -m unittest discover -s tests -v |
| Profile, plan, backlog | Add | none exist | create them with the CAN prefix |
| Agent instructions | Add | no CLAUDE.md or AGENTS.md | CLAUDE.md plus an AGENTS.md symlink |
| Design spec | Avoid | no UI | leave it out |
Riskiest assumption
The library has no technical risk; the unknown is whether the story flow runs end to end here. So
CAN-001 is a small real change pushed through the whole flow, with Done-when: bash scripts/verify-project.sh exits 0 on its branch.
After approval: setup committed to main. Setup validation passed and bash scripts/verify-project.sh passes on the existing tests. Files: CLAUDE.md, AGENTS.md, four docs (profile, plan, backlog, engineering guide), specs/_TEMPLATE.md, notes/_TEMPLATE.md, scripts/verify-project.sh. Nothing was overwritten.
Next steps
- Start the first story:
start story CAN-001 - Review the generated verifier commands before relying on them
Connects securely to your tools. The creator never sees your data.
About this skill
What it does
The problem. Agents start coding before anyone decided what the product is, what "done" means, or which assumption could sink it. The result is a backlog of vague tasks and a verifier nobody wrote.
What the skill does. It classifies the product (web, SaaS, mobile, API, CLI, SDK, data, AI and more), finds the riskiest assumption and puts it first in the backlog, and writes every Done-when item as a command or an artifact check. It generates the plan, backlog, engineering guide, unified repository instructions (AGENTS.md, CLAUDE.md, GEMINI.md), and scaffolds scripts/verify-project.sh from your stack markers.
Why not just prompt for it. The skill ships templates, a bootstrap script with a dry run, a validator, and a verifier scaffold that writes a failing TODO when it cannot recognize your stack, so an unconfigured project never reports a fake pass.
Use it when you start a greenfield product or adopt an existing repository for agent-assisted delivery across any AI coding tool or model.
Known limits.
- Software projects only.
- Generated verifier commands are a starting point; review them before relying on them.
Where it fits
- Project Setup (this skill), once per repository: plan, backlog, verifier
- Multi-Agent Story Flow or Dual-Session Story Flow, once per story: build, verify, independent review
- You: read the diff, push, open the PR, merge
- Merged Worktree Cleanup, after the merge: remove the worktree and branch
Project Setup is optional but recommended: without it the story flows fall back to auto-detected checks, with no backlog or project-owned verifier.
How to use
-
Unzip once into a shared folder that every agent can use:
mkdir -p ~/.agents/skills/project-setup && unzip -o project-setup-<version>.zip -d ~/.agents/skills/project-setup. Codex CLI and Gemini CLI read~/.agents/skillsdirectly. Point the others at the same folder instead of making a second copy:- Claude Code:
mkdir -p ~/.claude/skills && ln -sfn ~/.agents/skills/project-setup ~/.claude/skills/project-setup - Cursor:
mkdir -p ~/.cursor/skills && ln -sfn ~/.agents/skills/project-setup ~/.cursor/skills/project-setup - Antigravity:
mkdir -p ~/.gemini/config/skills && ln -sfn ~/.agents/skills/project-setup ~/.gemini/config/skills/project-setup
Keep only this one copy: a second install of the same skill can make the agent pick the wrong one.
- Claude Code:
-
Say
set up this projectorset up this repo for the story flow. -
Review the dry-run file list and approve it.
Requirements
python3 and bash. Works in any agent that loads SKILL.md skills, including Claude Code, Codex CLI, Cursor, Google Antigravity, Gemini CLI, Windsurf, GitHub Copilot, and OpenCode. With --instructions all, bootstrap_project.py writes AGENTS.md as the single source of truth plus CLAUDE.md and GEMINI.md that import it via @AGENTS.md.
Works best with: Multi-Agent Story Flow, Dual-Session Story Flow, and Merged Worktree Cleanup.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
This skill has been used 5 times.
Trust & safety
Security scan
Security scan passed 2 days ago
- Free with an Agensi account
Requires