- Home
- Skills
- DevOps & Cloud
- Project Bootstrap: CLAUDE.md, Rules, Skills & Infra
More screenshots
Works with the AI Agents you already use
Project Bootstrap: CLAUDE.md, Rules, Skills & Infra
Every new Claude Code session starts from zero: you re-explain the stack, the rules, and what's already decided.
Free
Project Bootstrap: CLAUDE.md, Rules, Skills & Infra
Example session with this skill installed
Run project-bootstrap for this Next.js project. We are a T1 solo team building a financial dashboard with PII data. Use conventional commits and trunk-based development.
- Read your context and instructions
- Compiled the project bootstrap
A real run on the docs from the project-docs demo (a surf forecast app), answering default at gate 1 and approve at gate 2. About 14 minutes start to finish. Below are the gate 2 assessment counts, then the skill's final report, unedited.
Gate 2: infrastructure assessment by category
| Category | ACTIVE | MANAGED | PLANNED | EXCLUDED | Total |
|---|---|---|---|---|---|
| Networking | 2 | 5 | 0 | 4 | 11 |
| Resilience | 3 | 0 | 0 | 4 | 7 |
| Messaging | 0 | 0 | 0 | 7 | 7 |
| Consistency | 2 | 1 | 0 | 6 | 9 |
| Database | 3 | 2 | 1 | 4 | 10 |
| Caching | 3 | 0 | 0 | 0 | 3 |
| Compute | 3 | 0 | 0 | 2 | 5 |
| Scaling | 0 | 2 | 0 | 1 | 3 |
| Deployment | 3 | 1 | 0 | 5 | 9 |
| Observability | 7 | 1 | 1 | 2 | 11 |
| Security | 4 | 3 | 1 | 5 | 13 |
| Reliability | 2 | 2 | 0 | 2 | 6 |
| Performance | 1 | 0 | 1 | 2 | 4 |
| Infrastructure | 1 | 0 | 0 | 2 | 3 |
| Operations | 4 | 0 | 0 | 1 | 5 |
| API, Deps, Network | 5 | 0 | 0 | 0 | 5 |
| Reference total | 43 | 17 | 4 | 47 | 111 |
| Client list | 17 | 0 | 4 | 5 | 26 |
| Grand total | 60 | 17 | 8 | 52 | 137 |
Final report
Bootstrap is finished. Claude Code sessions in this repo now start with the project's rules, status and workflows already loaded. All verification checks passed. I ran git init but committed nothing, and I didn't install the Expo plugin; that's one command for you, below.
Verification
CLAUDE.mdis 160 lines..claude/settings.jsonparses. The format hook exits cleanly on any input, and it formatted a test file correctly when Prettier was present.- Reading a test
.envwas denied, and the test file is deleted. - All 12 skill files have valid frontmatter. Every section number and doc path they mention exists.
- No secrets, banned words, dashes or curly quotes in any generated file.
- Cold-read test: an agent with no context, reading only
CLAUDE.mdandSTATUS.md, answered all seven questions correctly. It found eight gaps, mostly about how the first tasks work before any code or CI exists. I fixed all of them. - Not yet testable: the skills'
npmcommands and the Prettier hook. They need the scripts and dependencies that T-001 creates.
Infrastructure assessment (137 items, full detail in docs/INFRASTRUCTURE.md)
| ACTIVE | MANAGED | PLANNED | EXCLUDED | |
|---|---|---|---|---|
| Standard backend list (111) | 43 | 17 | 4 | 47 |
| Mobile client list (26) | 17 | 0 | 4 | 5 |
| Total | 60 | 17 | 8 | 52 |
Files created
CLAUDE.md: the rules every session loads; importsdocs/STATUS.md..claude/rules/:security.mdapplies everywhere.mobile.md,worker.md,core.md,testing.mdandinfra.mdload only when their folders are touched.
.claude/agents/reviewer.md: a read-only reviewer..claude/skills/: 12 skills (listed below)..claude/settings.json: allowed commands, file and push blocks, a format hook, and the Expo plugin enabled for the project..claude/hooks/format.mjs: runs Prettier on code files after each edit. It skips Markdown and never blocks..gitignore,.env.exampleandREADME.mdat the repo root.docs/STATUS.mdanddocs/INFRASTRUCTURE.md.
Files changed
docs/DECISIONS.md: added ADR-024 to ADR-028.- ADR-024: no feature flags.
- ADR-025: no certificate pinning.
- ADR-026: over-the-air updates deferred.
- ADR-027: no Docker or Terraform.
- ADR-028: development builds call the preview Worker over HTTPS.
docs/PLAN.md: new task T-048 (deploy and rollback runbook); Dependabot added to T-005.docs/README.md,docs/TECH_SPEC.md,docs/ARCHITECTURE.md: index rows, thecheck:sizescript, and the development-build change from ADR-028.
ADR-028 resolves one conflict I hit: local development builds would have needed unencrypted HTTP to reach a local Worker. They now call the preview Worker instead.
The five deferred items most likely to be needed first
- Over-the-air updates: the first app-only bug fix that can't wait for store review.
- Push notifications: when F-11 (window alerts) starts, which is first in line for v1.x.
- Deep links: needed by the same F-11 notifications so they can open a break.
- Error budgets: if bundle freshness misses its target two weeks running, or a second contributor joins.
- Firewall rules: if you add a custom domain, or the usage ping gets abused.
How to use it
- Install the plugin once:
claude plugin install expo@claude-plugins-official --scope project - Skills:
/next-task: picks, implements and verifies the next task in PLAN.md./handoff: updates STATUS.md and suggests commit messages./decision: records an ADR. Claude may run this one on its own; all the others only run when you call them./new-feature: specifies a feature not yet in the docs, then scaffolds it./rules-check: reviews your uncommitted changes againstCLAUDE.md./infra-audit,/security-audit,/perf-audit: check the infrastructure, security and performance targets against the actual code./scale-up: adds a deferred infrastructure item once its trigger fires./incident,/release,/bootstrap: incident records, release steps, and re-running this setup.
CLAUDE.local.md(not committed) holds personal preferences only, such as which test phone you have or local paths. Never team rules.
Suggested commit (this completes T-002; I haven't committed)
chore(repo): add project docs and Claude Code setup
Product and engineering docs (brief, PRD, SRS, UX, architecture,
tech spec, security, test plan, plan, decisions, launch, feature
specs F-01 to F-08, OpenAPI contract), infrastructure assessment
and status. CLAUDE.md, path-scoped rules, project skills, reviewer
agent, settings with a format hook, .gitignore and .env.example.
Next is T-001: set up the npm workspaces repo, tsconfig.base.json, ESLint and Prettier. Its first step installs development dependencies (TypeScript, ESLint, Prettier), so I'll show you the exact list before running anything. Want me to start it with /next-task, or commit first?
Connects securely to your tools. The creator never sees your data.
About this skill
Every new Claude Code session starts from zero: you re-explain the stack, the rules, and what's already decided. project-bootstrap reads your project docs and code once, then writes that knowledge into the repo, where every future session and every teammate who clones it picks it up. Works for new and existing projects, web or mobile.
What it does
- CLAUDE.md under 200 lines: role, stack with versions, workflow, hard boundaries, security, git, testing, performance budgets, layout and setup. Each rule cites its source ID from your docs (NFR-004, ADR-002).
- Path-scoped rules in
.claude/rules/that load only when matching files are touched: mobile, backend, database, tests, CI. - Project skills using this repo's real commands:
/next-task,/handoff,/decision,/new-feature,/rules-check,/infra-audit,/security-audit,/perf-audit,/scale-up,/incident,/release. Skills the project doesn't need are dropped. - A read-only reviewer subagent that checks a diff against your rules.
- Shared
.claude/settings.json: allow-listed lint, test and build commands, blocked reads of env and signing files, a format-on-edit hook. - Infrastructure assessment: 111 production concerns plus 26 web and mobile checks, each marked Active, Managed, Planned (with a measurable trigger) or Excluded (with a reason), saved to
docs/INFRASTRUCTURE.md. - Only the infrastructure your tier needs: CI on every PR and
.env.examplefor everyone; Docker, deploy workflows, IaC and tracing only at the tiers that need them.
How it works
- Detect: reads the docs from project-docs first, then manifests, CI, schema and any existing AI config (CLAUDE.md, AGENTS.md, .cursorrules), and reports every conflict between the docs and the code.
- Discover: asks only what the docs and code couldn't answer. With a full docs set, that's usually just the Claude Code setup. Confirmation gate 1.
- Assess: classifies all 137 items and shows the full plan: every file to create or change, every skill, every proposed plugin or dependency. Confirmation gate 2. Nothing is written before you approve it.
- Generate and verify: writes the files, validates every skill (and runs its read-only commands once the project has them),
checks settings.json parses, triggers the format hook, confirms
.envreads are blocked, greps for secrets, and runs a cold-read test: a fresh agent reads only CLAUDE.md and must answer how to run the tests, what's forbidden, and where the next task's spec is. - Report: counts, every file touched, the Planned triggers most likely to fire first, a suggested commit message, and an offer to start the first task in your plan.
Frameworks & tools
Stack-agnostic. It reads manifests for Node, Expo and React Native, Flutter, native iOS and Android, Go, Python, Rust, PHP, Ruby and .NET, plus GitHub Actions, GitLab CI, Docker, Vercel, Netlify, Cloudflare, Firebase and Supabase configs, and uses your real commands. If your team also uses Cursor, Codex or Copilot, shared instructions go in AGENTS.md, imported by CLAUDE.md.
Why this beats prompting it yourself
/init writes a CLAUDE.md from your code. This writes the whole team setup from your code and your decisions: rules tied to requirement IDs, workflows as skills, guardrails in settings, and infrastructure sized to your real tier, with a written reason for everything it leaves out, so future sessions stop suggesting Kubernetes for a solo MVP. It never overwrites an existing file. It proposes a merge and shows the diff.
Use cases
- Turning a fresh docs set into a ready-to-build repo, then starting the first task.
- Adding team-wide Claude Code context to an existing codebase without breaking working configs.
- Onboarding: teammates clone the repo and get the same rules, skills and guardrails.
- Staying current: the generated
/bootstrapskill re-checks the repo, and/scale-upmoves a Planned item to Active when its trigger fires.
Known limitations
- Requires Claude Code. Everything it writes is Claude Code configuration.
- Works best after project-docs. Without docs it recommends running that first and continues only if you agree.
- For a new project it can scaffold with the framework's official generator, which needs network access and installs dependencies. It asks first, and may leave scaffolding to the first tasks in your plan instead.
- Stops at two confirmation gates, and asks before git init, commits, pushes, and installing plugins, MCP servers or dependencies. Never commits.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for DevOps & Cloud.
Requires manual approval at two gates. Does not support direct cloud resource provisioning. Best results require running project-docs first. Restricted from reading .env values.
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 buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 19 days ago
- Free with an Agensi account
- Software
Node.jscreate-expo-appClaude Code
Git
Permissions
File Scopes