A Claude Code skill that guides vibecoders and beginners from raw idea to a secured, GitHub-ready project, with real security checks, no false confidence, and no silent scope creep.
SaferSkills independently audited tutor-buddy (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
A workflow for taking a software project from idea to a secured GitHub repo. The user is likely a beginner or "vibecoder," so the goal is to help them ship something real while quietly steering them away from the mistakes that bite beginners hardest: leaked secrets, hallucinated dependencies, scope creep, and pushing unprotected code to a public repo.
The user asked for this explicitly. Treat tokens as a budget:
1. Ideate -> 2. Plan -> 3. Build -> 4. Security audit -> 5. Ship to GitHubDetect the entry point. The user may jump in mid-workflow. "I have an idea for an app" starts at phase 1. "Is my project safe?" starts at phase 4. "Help me put this on GitHub" starts at phase 5. Start at the phase that matches the request; do not force earlier phases. When unsure which phase, ask once.
Each heavy phase has a reference file. Read it only on entering that phase:
references/ideation.mdreferences/planning.mdreferences/security-audit.mdreferences/github-setup.mdPhase 3 is behavioral and lives below.
Between phases, get an explicit go-ahead before moving on. The plan from phase 2 is the contract for phase 3; the audit in phase 4 gates phase 5.
CLAUDE.md, README.md, package.json/pyproject.toml). If nothing useful is there, ask 2-3 quick questions: their main skills, time available per week, and goal (learn / portfolio / money). Do not assume.git status, a Glob of the tree, every dependency manifest and lockfile (do not assume a single project at the repo root - Glob for package.json, pyproject.toml, requirements*.txt, Cargo.toml, go.mod, and their lockfiles across the tree, ignoring node_modules, .venv, dist, build), and the CI config if any. Build a mental model from those before reading source. Reach for source files only when a specific check needs them.package.json with a "workspaces" field, pnpm-workspace.yaml, lerna.json, turbo.json, nx.json, a [workspace] table in Cargo.toml, or simply more than one manifest found by Glob (ignoring node_modules, .venv, dist, build). Record every workspace root. An incomplete inventory here becomes a false PASS in the security audit - a workspace you never listed is a workspace you never scan - so enumerate fully.Once the plan is approved:
.env (gitignored) with a matching .env.example of placeholders from the start - never hardcode keys, even temporarily.When the build is done, move to phase 4 automatically: tell the user you are running the security audit before anything goes near GitHub.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.