Built for the 2026 JS, Next.js & AI-agent threat surface

AppSec · OpSec · Cyber Defense

Erydas Supply-chain and live-app defense that never invents findings.

2026 brought self-propagating npm worms, 50+ Next.js advisories, and poisoned AI agent configs. Erydas gives builders and security teams one deterministic shield: exact lockfile truth from OSV, passive L1–L4 live-app recon, and verified PR-only fixes with lifecycle scripts disabled.

Advisory truth from . Optional AI explains confirmed findings only — never invents CVEs or picks versions.

  • Read-only firstuntil you opt into a PR
  • Scripts disabled--ignore-scripts on lockfile apply
  • Fail closedno auto-merge, no silent guesses
$npx eridas scan .

OpSec recon

Passive · GET-only · <1 min

Scan a live app

Grade any public URL from headers and AI-crawler policy to leaked .env, .git, source maps, and secrets in JS bundles. Free, no account needed.

  • CSP & HSTS
  • AI-crawler rules
  • .env & .git leaks
  • JS bundle secrets
  • OSV packages

Lockfile inventory stays local: npx eridas scan .

2026 Threat Briefing

Why 2026 broke legacy scanners — and what replaces them.

Modern JS, Next.js, and AI-assisted teams are being hit across lockfiles, live edges, and coding-agent configs at once. Erydas pairs deterministic OSV truth with passive recon and hard execution boundaries.

  1. 2026 Supply ChainOSV · MAL-* · --ignore-scripts

    Self-propagating npm worms & postinstall droppers

    Poisoned releases across mainstream npm packages (Axios 1.14.1, Mastra, and the ChainDrop worm hitting 1,300+ versions including keyv and flat-cache) ran preinstall/postinstall payloads to harvest .env files, GitHub PATs, and cloud keys.

    How Erydas stops the blast radius

    Queries exact package-lock.json versions against OSV — surfacing MAL-* malware records alongside CVEs — and regenerates lockfiles with --ignore-scripts and atomic rollback.

    $ npx eridas scan . --severity low
  2. 2026 Framework WaveNext.js · Edge · L1–L3 Recon

    50+ Next.js advisories & live-edge exposure drift

    Framework advisory velocity surged in 2026 — spanning unauthenticated RCEs in cache and image pipelines, cross-tenant cache disclosure, cookie exposure, and leaked source maps or .env routes in production.

    How Erydas verifies code + edge

    Pairs lockfile-to-OSV checks with passive GET-only L1–L4 recon: CSP, HSTS, CORS, AI-crawler policy, leaked .env/.git, source maps, and secrets embedded in public JS bundles.

    $ npx eridas tinfoil https://your-app.com --level 3
  3. 2026 Agentic SecurityOWASP MCP · Zero Hallucinations

    AI coding agents & MCP servers as a new attack surface

    Poisoned .cursorrules, CLAUDE.md, and MCP tool descriptions turned helpful coding assistants into confused deputies that exfiltrate secrets or execute unreviewed shell commands.

    How Erydas keeps agents inside rails

    Treats repo text, PR bodies, and advisory prose as untrusted input. Optional AI explains confirmed hits only — it never invents CVEs, picks versions, widens a scope grant, or merges to main.

    $ npx eridas fix . --dry-run --quiet --format json
  4. 2026 Org GovernanceTrusted Execution · PR-Only

    Shipping at AI speed without handing keys to CI bots

    GitHub Actions pwn requests and over-scoped bot tokens turned automated dependency workflows into supply-chain entry points across entire organizations.

    How Erydas separates control & execution

    Read-only scan mode (contents:read), dedicated PR branches (eridas/automated-remediation), tenant-isolated Supabase RLS, and customer-owned no-egress connectors for experimental proofs.

    $ npx eridas init . --mode scan

Every control runs identically on your laptop, in GitHub Actions, in Erydas Cloud, and over @eridas/mcp.

Built for your stack

Solo builders and enterprise defenders, on the same tool.

Whether you ship one Next.js app with Cursor or defend fifty repositories across GitHub, Vercel, and Supabase — Erydas gives you deterministic answers with zero lock-in.

Zero account · Local CLI · Cursor & Claude Code

Ship Next.js, Vite, and Node apps without getting burned on launch day.

Run one command before you deploy or let your coding agent call the exact same deterministic tool over MCP — no SaaS bloat, no opaque scores, no noisy false positives.

  • npm
  • Next.js
  • React
  • Vite
  • Expo
  • Node.js
  • TypeScript
  • Cursor
  • Claude Code
  • Codex
$npx eridas scan . && npx eridas tinfoil https://your-app.com --level 2
  1. 01

    60-second pre-launch recon

    Paste your Vercel or production URL to grade headers, CSP, AI-crawler rules, leaked .env/.git, source maps, and exposed bundle secrets.

  2. 02

    Exact lockfile hygiene, zero guesswork

    eridas scan checks every direct and transitive version in package-lock.json against OSV; eridas fix plans the smallest bump with --ignore-scripts.

  3. 03

    Drop-in Agent Skill & MCP server

    Install SKILL.md or @eridas/mcp so Cursor, Claude Code, and Codex check advisories deterministically instead of hallucinating package versions.

For the people on the watch

Built for cyber defence, not a findings feed.

Erydas uses the same disclosed OSV records the community already cites. It tells you what it may do, and it stops where a defender would stop: no invented CVE, no silent push, no experiment that can publish.

  1. AppSec

    The lockfile is the inventory

    Exact installed versions, disclosed OSV records, and the smallest direct bump. pnpm and Yarn are refused instead of approximated. Transitive findings stay visible and are not overridden.

  2. OpSec

    The live edge stays passive

    A GET-only recon of a public URL. Private, loopback, and metadata addresses are refused. Level 2 is the default: headers, leaked files, and secrets in bundles.

  3. Cyber defence

    Review is a pull request

    Connect GitHub, then choose scan-on-push, pull-request review, and whether a fix branch opens itself. Erydas never merges to the default branch.

How you engage

Three steps. Stop after any of them.

Look never changes the tree. Propose opens a pull request and waits. Prove runs only inside a disposable environment that cannot publish.

01 · Look

See what is installed and what is exposed

You do

  1. Run eridas scan on a lockfile, or paste a public URL in the Scan workspace.
  2. Choose a level. Level 2 is the default live scan: headers plus leaked files and secrets in bundles.
  3. Read the report. Exit 1 means disclosed findings remain. That is not a clean tree.

Erydas may

  • Inventory package-lock.json lockfile versions 2 and 3.
  • Query OSV for those exact versions.
  • Run a passive GET-only recon of a public URL.
  • Store the scan and show it beside older runs.

Hard stop

  • Does not run npm install or lifecycle scripts.
  • Does not rewrite package.json or the lockfile.
  • Does not delete findings for the repository.
  • Refuses private, loopback, and metadata addresses.
  • Refuses pnpm and Yarn instead of guessing their trees.
npx eridas scan . --quiet --format jsonOpen Scan

Always in force

  • OSV decides what is disclosed. Erydas does not invent advisories.
  • package-lock.json decides what is installed. A stale lockfile is an error.
  • A person decides whether anything changes. There is no silent push to the default branch.
  • Older findings stay. The current view is the latest succeeded scan.
  • One organization cannot read another organization's jobs, events, or grants.

Who it is for

Built for the people who have to sign the risk.

AppSec closes disclosed dependency vulns. OpSec sees the live attack surface. Cyber and AI teams run the same contracts in CI, Cursor, and review. One scanner. Zero invented findings.

  • AppSec

    Close disclosed dependency risk

    Treat package-lock.json as inventory. Plan the minimum direct bump that closes every advisory. Prove it with a rescan before anyone merges.

    • Exact installed versions — direct and transitive
    • Surfaces CVE, GHSA, and MAL-* malware disclosures
    • Lifecycle scripts disabled while the lockfile regenerates
    • Manifest and lockfile restore together if verification fails
  • OpSec

    See what the live product leaks

    A passive, GET-only recon of the running app. Four cumulative levels, one graded report. Private, loopback, and metadata addresses are refused.

    • L1 Surface — headers, CSP, HSTS, CORS, AI crawlers, llms.txt
    • L2 Exposure — .env, .git, source maps, secrets in bundles
    • L3 Supply chain — OSV on announced packages or a lockfile
    • L4 Reconstruction — what an attacker's agent can rebuild (local)
  • Cyber

    Same truth in CI, SOAR, and coding agents

    Humans, pipelines, and AI coding agents share one contract: stable JSON, exit codes, dry-run, and PR-only mutation. Never a silent push to default.

    • Exit 0 clean · 1 findings remain · 2 operational · 3 declined
    • schemaVersion: 1 JSON on stdout; progress on stderr
    • Cloud API, TypeScript SDK, @eridas/mcp, and portable SKILL.md
    • Remediation PRs on eridas/automated-remediation

Capabilities

What Erydas actually does.

Not a CVE chatbot. Not an auto-merge bot. Six capabilities with the same trust boundary: OSV decides what is disclosed, the lockfile decides what is installed, and a human or an explicit flag decides whether anything mutates.

  1. 01

    AppSec

    Lockfile inventory, not a guess

    Erydas reads package-lock.json (lockfileVersion 2 or 3) and inventories every exact installed version. A stale lockfile is an error. pnpm and Yarn are refused — never silently approximated.

    npx eridas scan . --quiet --format json
    scan
  2. 02

    AppSec

    OSV is the only advisory truth

    Exact (ecosystem, name, version) queries against OSV. GHSA, CVE aliases, and MAL malware IDs collapse to one finding. Unknown severity stays visible. AI may narrate a confirmed hit — it cannot invent a CVE or pick a version.

    npx eridas scan . --severity high
    Why OSV
  3. 03

    OpSec

    Live-app exposure, four levels deep

    Tinfoil is a passive recon of the running product: surface, exposure, supply chain, reconstruction. Same report from the Scan workspace, CLI, Cloud API, SDK, and MCP. Default is Level 2.

    npx eridas tinfoil https://example.com --level 2
    Levels
  4. 04

    AppSec

    Plan, approve, apply, prove

    fix prints a plan first. Non-interactive sessions cannot mutate without --yes or --force. --force skips the prompt only — not major policy, ignore-scripts, OSV rescan, verificationCommands, or rollback.

    npx eridas fix . --dry-run --quiet --format json
    Pipeline
  5. 05

    Cyber

    Remediation is a pull request

    init --mode fix and Cloud remediations land on eridas/automated-remediation for normal branch protection. Erydas never auto-merges to default. Repeated runs update the same branch and PR.

    npx eridas init . --mode fix
    Runbook
  6. 06

    Cyber

    CI, SOAR, and agent contracts

    Gate jobs on exit codes, not log text. Drive the same pipeline from GitHub Actions, /api/v1, @eridas/cloud-sdk, or @eridas/mcp. API keys live in secrets — never in eridas.config.json.

    eridas cloud scan --repo-id <uuid> --wait --format json
    API

Source of truth

Every Erydas finding starts on OSV.

Open Source Vulnerabilities is the open, precise, distributed database for disclosed advisories — including CVEs, GHSAs, and MAL supply-chain malware records. Erydas matches your lockfile to OSV, then plans the smallest verified fix. No invented CVEs. No opaque scores.

  • Open

    Advisories from authoritative open sources — GitHub, OpenSSF Malicious Packages, PyPA, RustSec, and more.

  • Precise

    Machine-readable affected ranges that map to exact package versions and commits.

  • Actionable

    Erydas turns OSV disclosure into a minimum safe bump — then rescans to prove it.

Version control

Connect GitHub. Scan, fix, merge.

Link repos once through the GitHub App. Erydas Cloud scans your lockfile, opens deterministic fix PRs in an isolated worktree, and tracks the same PR lifecycle you see in the dashboard — queued, running, open, merged.

  1. ConnectInstall the Erydas GitHub App on your org or repo.
  2. ScanQueued → running → succeeded against your lockfile.
  3. FixDeterministic bump applied in an isolated worktree branch.
  4. PR openAutomated PR opened for review on GitHub.
  5. MergedMerge on GitHub — worktree returns to default branch.

Same contracts for the dashboard, /api/v1, SDK, CLI, and MCP — sync repos, poll scan and remediation status, read git worktree state.

On a real repository

Exact eridas scan output — a read-only preview of what Eridas Cloud shows after syncing a repo.

PreviewLogged-out · scan snapshot onlySign in to remediate →
succeeded

Dependencies

332

Findings

35

Urgent

14

Command

npx eridas scan .

Severity breakdown

35 disclosed
crit1high13mode16low5

eridas — zsh

e20a7e1
Eridas scanned 332 dependencies in .
source: OSV (https://osv.dev)
result: 35 disclosed vulnerability finding(s)
summary: 1 critical, 13 high, 16 moderate, 5 low
 
[CRITICAL] GHSA-5xrq-8626-4rwp: When Vitest UI server is listening, arbitrary file can be read and executed
package: vitest@2.1.9 (direct)
path: vitest
fixed: 3.2.6, 4.1.0
reference: https://github.com/vitest-dev/vitest/security/advisories/GHSA-5xrq-8626-4rwp
 
[HIGH] GHSA-25h7-pfq9-p65f: flatted vulnerable to unbounded recursion DoS in parse() revive phase
package: flatted@3.3.3 (transitive)
path: eslint > file-entry-cache > flat-cache > flatted
fixed: 3.4.0
reference: https://github.com/WebReflection/flatted/security/advisories/GHSA-25h7-pfq9-p65f
 
[HIGH] GHSA-rf6f-7fwh-wjgh: Prototype Pollution via parse() in NodeJS flatted
package: flatted@3.3.3 (transitive)
path: eslint > file-entry-cache > flat-cache > flatted
fixed: 3.4.2
reference: https://github.com/WebReflection/flatted/security/advisories/GHSA-rf6f-7fwh-wjgh
 
[HIGH] GHSA-hmw2-7cc7-3qxx: form-data: CRLF injection in form-data via unescaped multipart field names and filenames
package: form-data@4.0.4 (transitive)
path: jsdom > form-data
fixed: 4.0.6
reference: https://github.com/form-data/form-data/security/advisories/GHSA-hmw2-7cc7-3qxx
 
[HIGH] GHSA-r5fr-rjxr-66jc: lodash vulnerable to Code Injection via `_.template` imports key names
package: lodash@4.17.21 (transitive)
path: @testing-library/jest-dom > lodash
fixed: 4.18.0
reference: https://github.com/lodash/lodash/security/advisories/GHSA-r5fr-rjxr-66jc
 
[HIGH] GHSA-23c5-xmqv-rm74: minimatch ReDoS: nested *() extglobs generate catastrophically backtracking regular expressions
package: minimatch@3.1.2 (transitive)
path: eslint > minimatch
paths: 3 dependency paths (showing first)
fixed: 3.1.4, 4.2.5, 5.1.8, 6.2.2, 7.4.8, 8.0.6, 9.0.7, 10.2.3
reference: https://github.com/isaacs/minimatch/security/advisories/GHSA-23c5-xmqv-rm74
 
[HIGH] GHSA-3ppc-4f35-3m26: minimatch has a ReDoS via repeated wildcards with non-matching literal in pattern
package: minimatch@3.1.2 (transitive)
path: eslint > minimatch
paths: 3 dependency paths (showing first)
fixed: 3.1.3, 4.2.4, 5.1.7, 6.2.1, 7.4.7, 8.0.5, 9.0.6, 10.2.1
reference: https://github.com/isaacs/minimatch/security/advisories/GHSA-3ppc-4f35-3m26
 
[HIGH] GHSA-7r86-cg39-jmmj: minimatch has ReDoS: matchOne() combinatorial backtracking via multiple non-adjacent GLOBSTAR segments
package: minimatch@3.1.2 (transitive)
path: eslint > minimatch
paths: 3 dependency paths (showing first)
fixed: 3.1.3, 4.2.5, 5.1.8, 6.2.2, 7.4.8, 8.0.6, 9.0.7, 10.2.3
reference: https://github.com/isaacs/minimatch/security/advisories/GHSA-7r86-cg39-jmmj
 
[HIGH] GHSA-c2c7-rcm5-vvqj: Picomatch has a ReDoS vulnerability via extglob quantifiers
package: picomatch@4.0.3 (transitive)
path: vite > picomatch
paths: 2 dependency paths (showing first)
fixed: 4.0.4
reference: https://github.com/micromatch/picomatch/security/advisories/GHSA-c2c7-rcm5-vvqj
 
[HIGH] GHSA-mw96-cpmx-2vgc: Rollup 4 has Arbitrary File Write via Path Traversal
package: rollup@4.45.1 (transitive)
path: vite > rollup
paths: 3 dependency paths (showing first)
fixed: 4.59.0
reference: https://github.com/rollup/rollup/security/advisories/GHSA-mw96-cpmx-2vgc
 
[HIGH] GHSA-fx2h-pf6j-xcff: vite: `server.fs.deny` bypass on Windows alternate paths
package: vite@5.4.19 (transitive)
path: vitest > vite-node > vite
paths: 2 dependency paths (showing first)
fixed: 6.4.3, 7.3.5, 8.0.16
reference: https://github.com/vitejs/vite/security/advisories/GHSA-fx2h-pf6j-xcff
 
[HIGH] GHSA-fx2h-pf6j-xcff: vite: `server.fs.deny` bypass on Windows alternate paths
package: vite@7.0.5 (direct)
path: vite
fixed: 7.3.5, 8.0.16
reference: https://github.com/vitejs/vite/security/advisories/GHSA-fx2h-pf6j-xcff
 
[HIGH] GHSA-p9ff-h696-f583: Vite Vulnerable to Arbitrary File Read via Vite Dev Server WebSocket
package: vite@7.0.5 (direct)
path: vite
fixed: 7.3.2, 8.0.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-p9ff-h696-f583
 
[HIGH] GHSA-96hv-2xvq-fx4p: ws: Memory exhaustion DoS from tiny fragments and data chunks
package: ws@8.18.3 (transitive)
path: jsdom > ws
fixed: 8.21.0
reference: https://github.com/websockets/ws/security/advisories/GHSA-96hv-2xvq-fx4p
 
[MODERATE] GHSA-2g4f-4pwh-qvx6: ajv has ReDoS when using `$data` option
package: ajv@6.12.6 (transitive)
path: eslint > ajv
paths: 2 dependency paths (showing first)
fixed: 6.14.0, 8.18.0
reference: https://nvd.nist.gov/vuln/detail/CVE-2025-69873
 
[MODERATE] GHSA-f886-m6hf-6m8v: brace-expansion: Zero-step sequence causes process hang and memory exhaustion
package: brace-expansion@1.1.12 (transitive)
path: eslint > minimatch > brace-expansion
paths: 3 dependency paths (showing first)
fixed: 1.1.13, 2.0.3, 3.0.2, 5.0.5
reference: https://github.com/juliangruber/brace-expansion/security/advisories/GHSA-f886-m6hf-6m8v
 
[MODERATE] GHSA-67mh-4wv8-2f99: esbuild enables any website to send any requests to the development server and read the response
package: esbuild@0.21.5 (transitive)
path: vitest > vite-node > vite > esbuild
paths: 2 dependency paths (showing first)
fixed: 0.25.0
reference: https://github.com/evanw/esbuild/security/advisories/GHSA-67mh-4wv8-2f99
 
[MODERATE] GHSA-h67p-54hq-rp68: JS-YAML: Quadratic-complexity DoS in merge key handling via repeated aliases
package: js-yaml@4.1.0 (transitive)
path: eslint > @eslint/eslintrc > js-yaml
fixed: 4.2.0
reference: https://github.com/nodeca/js-yaml/security/advisories/GHSA-h67p-54hq-rp68
 
[MODERATE] GHSA-mh29-5h37-fv8m: js-yaml has prototype pollution in merge (<<)
package: js-yaml@4.1.0 (transitive)
path: eslint > @eslint/eslintrc > js-yaml
fixed: 4.1.1
reference: https://github.com/nodeca/js-yaml/security/advisories/GHSA-mh29-5h37-fv8m
 
[MODERATE] GHSA-f23m-r3pf-42rh: lodash vulnerable to Prototype Pollution via array path bypass in `_.unset` and `_.omit`
package: lodash@4.17.21 (transitive)
path: @testing-library/jest-dom > lodash
fixed: 4.18.0
reference: https://github.com/lodash/lodash/security/advisories/GHSA-f23m-r3pf-42rh
 
[MODERATE] GHSA-xxjr-mmjv-4gpg: Lodash has Prototype Pollution Vulnerability in `_.unset` and `_.omit` functions
package: lodash@4.17.21 (transitive)
path: @testing-library/jest-dom > lodash
fixed: 4.17.23
reference: https://github.com/lodash/lodash/security/advisories/GHSA-xxjr-mmjv-4gpg
 
[MODERATE] GHSA-3v7f-55p6-f55p: Picomatch: Method Injection in POSIX Character Classes causes incorrect Glob Matching
package: picomatch@4.0.3 (transitive)
path: vite > picomatch
paths: 2 dependency paths (showing first)
fixed: 4.0.4
reference: https://github.com/micromatch/picomatch/security/advisories/GHSA-3v7f-55p6-f55p
 
[MODERATE] GHSA-qx2v-qp2m-jg93: PostCSS has XSS via Unescaped </style> in its CSS Stringify Output
package: postcss@8.5.6 (transitive)
path: vite > postcss
paths: 3 dependency paths (showing first)
fixed: 8.5.10
reference: https://github.com/postcss/postcss/security/advisories/GHSA-qx2v-qp2m-jg93
 
[MODERATE] GHSA-4w7w-66w2-5vf9: Vite Vulnerable to Path Traversal in Optimized Deps `.map` Handling
package: vite@5.4.19 (transitive)
path: vitest > vite-node > vite
paths: 2 dependency paths (showing first)
fixed: 6.4.2, 7.3.2, 8.0.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-4w7w-66w2-5vf9
 
[MODERATE] GHSA-4w7w-66w2-5vf9: Vite Vulnerable to Path Traversal in Optimized Deps `.map` Handling
package: vite@7.0.5 (direct)
path: vite
fixed: 7.3.2, 8.0.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-4w7w-66w2-5vf9
 
[MODERATE] GHSA-93m4-6634-74q7: vite allows server.fs.deny bypass via backslash on Windows
package: vite@5.4.19 (transitive)
path: vitest > vite-node > vite
paths: 2 dependency paths (showing first)
fixed: 5.4.21, 6.4.1, 7.0.8, 7.1.11
reference: https://github.com/vitejs/vite/security/advisories/GHSA-93m4-6634-74q7
 
[MODERATE] GHSA-93m4-6634-74q7: vite allows server.fs.deny bypass via backslash on Windows
package: vite@7.0.5 (direct)
path: vite
fixed: 7.0.8, 7.1.11
reference: https://github.com/vitejs/vite/security/advisories/GHSA-93m4-6634-74q7
 
[MODERATE] GHSA-v6wh-96g9-6wx3: launch-editor: NTLMv2 hash disclosure via UNC path handling on Windows
package: vite@5.4.19 (transitive)
path: vitest > vite-node > vite
paths: 2 dependency paths (showing first)
fixed: 6.4.3, 7.3.5, 8.0.16
reference: https://github.com/vitejs/launch-editor/security/advisories/GHSA-v6wh-96g9-6wx3
 
[MODERATE] GHSA-v6wh-96g9-6wx3: launch-editor: NTLMv2 hash disclosure via UNC path handling on Windows
package: vite@7.0.5 (direct)
path: vite
fixed: 7.3.5, 8.0.16
reference: https://github.com/vitejs/launch-editor/security/advisories/GHSA-v6wh-96g9-6wx3
 
[MODERATE] GHSA-58qx-3vcg-4xpx: ws: Uninitialized memory disclosure
package: ws@8.18.3 (transitive)
path: jsdom > ws
fixed: 8.20.1
reference: https://github.com/websockets/ws/security/advisories/GHSA-58qx-3vcg-4xpx
 
[LOW] GHSA-4x5r-pxfx-6jf8: @babel/core: Arbitrary File Read via sourceMappingURL Comment
package: @babel/core@7.28.0 (transitive)
path: @vitejs/plugin-react > @babel/core
fixed: 7.29.6, 8.0.0-rc.6
reference: https://github.com/babel/babel/security/advisories/GHSA-4x5r-pxfx-6jf8
 
[LOW] GHSA-g4jq-h2w9-997c: Vite middleware may serve files starting with the same name with the public directory
package: vite@5.4.19 (transitive)
path: vitest > vite-node > vite
paths: 2 dependency paths (showing first)
fixed: 5.4.20, 6.3.6, 7.0.7, 7.1.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-g4jq-h2w9-997c
 
[LOW] GHSA-g4jq-h2w9-997c: Vite middleware may serve files starting with the same name with the public directory
package: vite@7.0.5 (direct)
path: vite
fixed: 7.0.7, 7.1.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-g4jq-h2w9-997c
 
[LOW] GHSA-jqfw-vq24-v9c3: Vite's `server.fs` settings were not applied to HTML files
package: vite@5.4.19 (transitive)
path: vitest > vite-node > vite
paths: 2 dependency paths (showing first)
fixed: 5.4.20, 6.3.6, 7.0.7, 7.1.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-jqfw-vq24-v9c3
 
[LOW] GHSA-jqfw-vq24-v9c3: Vite's `server.fs` settings were not applied to HTML files
package: vite@7.0.5 (direct)
path: vite
fixed: 7.0.7, 7.1.5
reference: https://github.com/vitejs/vite/security/advisories/GHSA-jqfw-vq24-v9c3

Operators

How a security team turns it on.

Start read-only. Gate CI on exit codes. Scan the live app. Connect GitHub when you are ready for pull-request remediation — never a push to default.

  1. 01Baseline the lockfile (read-only)
  2. 02Schedule a read-only watch
  3. 03Scan the live product
  4. 04Connect GitHub for the org
  5. 05Remediate on a pull request
  6. 06Pin policy and prove the fix

Questions defenders ask

Straight answers, no marketing fog.

How Erydas treats OSV truth, npm lifecycle scripts, passive production scans, AI coding agents, and branch protection.

Why does Erydas use OSV instead of an AI vulnerability scanner?

LLMs hallucinate CVE IDs and invent non-existent package versions. Erydas uses Google's Open Source Vulnerabilities (OSV) database as the sole source of advisory truth — covering CVEs, GitHub Security Advisories (GHSAs), and MAL-* supply-chain malware records against exact installed versions in package-lock.json. Optional AI explains confirmed hits only.

How does Erydas protect against npm postinstall worms like ChainDrop or Axios?

Two ways: first, exact lockfile scanning surfaces disclosed OSV MAL-* and GHSA records across both direct and transitive dependencies; second, whenever Erydas applies a remediation and regenerates package-lock.json, it runs npm with --ignore-scripts enabled and rolls back both package.json and the lockfile atomically if rescan or verification fails.

Is the live URL scanner (Tinfoil) safe to run against production apps?

Yes. Tinfoil is strictly passive and GET-only — it sends no mutating payloads, performs no credentialed fuzzing, and refuses private, loopback, and cloud-metadata IP ranges to prevent SSRF. Level 2 checks headers, CSP, CORS, robots.txt AI-crawler rules, leaked .env/.git, source maps, and secrets in public JS bundles.

How does Erydas work with Cursor, Claude Code, Codex, and MCP?

Erydas ships a portable Agent Skill (/skills/eridas/SKILL.md), machine docs (/llms.txt and /agents.txt), and a stdio MCP server (@eridas/mcp). Agents invoke the exact same deterministic JSON contracts, dry-run preview, and exit codes (0/1/2/3) that humans and CI use — and remediation always requires explicit confirmation.

Will Erydas ever push directly to main or auto-merge a dependency bump?

Never. Scheduled workflows in scan mode request contents:read only. Fix mode and Erydas Cloud open or update a pull request on a dedicated branch (eridas/automated-remediation) so your existing CI checks and branch protection rules always have the final word.

Which frameworks and package managers does Erydas support today?

Erydas works with any npm project using package-lock.json (lockfileVersion 2 or 3) — including Next.js, React, Vite, Expo, Node.js, TypeScript, and monorepos — plus any public web application for live URL recon. pnpm and Yarn lockfiles are explicitly detected and refused rather than silently approximated.