# Erydas > Deterministic cyber defense for npm, live web apps, and AI agents. Erydas is open-source, deterministic supply-chain and live-app security built for solo builders, AppSec, OpSec, cyber defense teams, CI pipelines, and AI coding agents shipping on npm, Next.js, React, Vite, Expo, Node.js, TypeScript, Supabase, Vercel, and GitHub. It inventories exact installed versions from `package-lock.json`, matches them to disclosed OSV advisories (CVE, GHSA, and `MAL-*` malware records), runs passive GET-only L1–L4 live-app exposure recon, and applies the smallest verified direct-dependency update with `--ignore-scripts` and atomic rollback — through a pull request, never a push to default. Optional AI explains confirmed findings only; it never invents CVEs or picks versions. Vulnerability source of truth: https://osv.dev Repository: https://github.com/rcdev714/erydas Cloud: https://erydas.vercel.app ## Brand / audience - Name: Erydas (CLI & npm packages: `eridas`, `@eridas/cloud-sdk`, `@eridas/mcp`) - Tagline: Deterministic cyber defense for npm, live web apps, and AI agents. - Audience: solo builders, security and cyber defense teams, engineers, CI/CD, and AI coding agents (Cursor, Claude Code, Codex). - Open-source `eridas` CLI works with no account. - Erydas Cloud adds GitHub App org sync, Security Command Overview, `/api/v1` + `/api/v2`, `@eridas/cloud-sdk`, `@eridas/mcp`, and trusted-execution investigations. - Live-app recon: `eridas tinfoil` / `POST /api/v1/tinfoil` (levels 1–4, passive GET-only). - Same advisory truth in Guided UI, Dev UI, CLI, API, SDK, and MCP. - Operator runbook: /for-security-teams ## For humans - Home: / - How you engage (Look / Propose / Prove): /engage - Docs & Cloud API: /docs - For security and cyber teams: /for-security-teams - Why OSV: /osv - Scan levels (L1–L4): /scan-levels - How it works: /how-it-works - Commands: /commands - Configuration: /configuration - Security model: /security - Architecture: /architecture - For agents & MCP: /agents - Extending: /extending ## For agents Erydas is easy for agents because the product is deterministic: stable JSON, exit codes (0/1/2/3), dry-run, and non-interactive flags. Prefer those over prompts. Never invent advisories. OSV is the only vulnerability source of truth. ### Discovery files - Agent index: /agents.txt - This file: /llms.txt - Portable skill: /skills/eridas/SKILL.md - Agents guide: /agents - Docs hub: /docs - Security team runbook: /for-security-teams - Commands: /commands ### Portable Agent Skill (any coding agent) Download: https://erydas.vercel.app/skills/eridas/SKILL.md ```sh mkdir -p .agents/skills/eridas && curl -fsSL https://erydas.vercel.app/skills/eridas/SKILL.md -o .agents/skills/eridas/SKILL.md ``` Also works under `.cursor/skills/eridas/`, `.claude/skills/eridas/`, or `.codex/skills/eridas/` — same file. ### Open-source CLI (no account) ```sh npx eridas scan . npx eridas scan . --quiet --format json npx eridas scan . --ai npx eridas fix . --dry-run --format json npx eridas fix . --yes npx eridas init . --mode fix ``` ### Erydas Cloud API (GitHub-connected org) ```sh export ERIDAS_API_KEY=erd_… export ERIDAS_API_URL=https://erydas.vercel.app eridas auth login eridas auth whoami --format json eridas cloud repos --format json eridas cloud scan --repo-id --wait --format json eridas cloud findings --repo-id --format json eridas cloud remediate --repo-id --wait --format json ``` Health endpoint: `GET /api/health` (`?strict=1` returns 503 when degraded). SDK: `@eridas/cloud-sdk` — typed client for `/api/v1` (`me`, repos, sync, git-status, package-root, scans, findings, remediations, tinfoil, wait helpers). MCP: `@eridas/mcp` — Cursor/IDE tools (local + Cloud). Set `ERIDAS_API_KEY` in MCP env. ### Scan a live site by level (Tinfoil) One report with a grade per level; each level includes the ones below. 1 Surface, 2 Exposure (default), 3 Supply chain, 4 Reconstruction. - Web page (labelled form, Enter submits, deep links start on their own): `/scan?url=https://example.com&level=2` and `/scan?level=3&packages=lodash@4.17.20,minimist@1.2.0` - Saved report permalink: `/scan?scan=` - API: `POST /api/v1/tinfoil` with `{"url":"https://example.com","level":2}` (Bearer `ERIDAS_API_KEY`), `GET /api/v1/tinfoil`, `GET /api/v1/tinfoil/:id` - CLI: `npx eridas tinfoil https://example.com --level 2 --format json` - SDK: `client.tinfoilScan({ url, level })` · MCP: `eridas_tinfoil_scan` (local), `eridas_cloud_tinfoil_scan` (Cloud) - Private, loopback and cloud-metadata addresses are refused. Scans are passive, GET-only. ### Exit codes - 0: clean / success - 1: disclosed findings remain - 2: operational / configuration error - 3: interactive fix approval declined ### JSON contracts - `eridas scan --format json` → `ScanResult` (`schemaVersion: 1`) - `eridas fix --format json` → single `FixResult` envelope (`kind: "fix"`, `status`, `plan`, `scan`) - Progress goes to stderr; use `--quiet` in CI pipelines ### Rules agents must follow 1. Query OSV via Erydas; do not scrape forums or invent CVEs. 2. Use `--dry-run` before mutating unless the user explicitly requested apply. 3. `--force` / `--yes` skip prompts only; they do not allow major upgrades without `--allow-major` and never disable `--ignore-scripts` or rollback. 4. Optional `--ai` explains confirmed findings only; it must not select versions. 5. Prefer opening/updating a remediation PR over pushing to the default branch. 6. Cloud GitHub-gated endpoints need the App connected; otherwise expect `github_not_connected`. ### Key links - OSV: https://osv.dev - OSV API: https://google.github.io/osv.dev/api/ - OSV schema: https://github.com/ossf/osv-schema - GitHub: https://github.com/rcdev714/erydas - Cloud: https://erydas.vercel.app - Agent index: /agents.txt - This file: /llms.txt - Agent skill: /skills/eridas/SKILL.md - Docs hub: /docs - For security teams: /for-security-teams - Agents: /agents