How it works

Read the lockfile, ask OSV about exact versions, plan the smallest safe bump, get approval, apply with scripts disabled, then rescan and verify. Same pipeline for the open-source CLI, Cloud API/SDK, Guided UI, Dev mode, CI, and agents.

Choose a step first on How you engage. Look is read-only, Propose is a pull request, and Prove cannot publish.

Security and cyber teams: start with the operator runbook.

Lockfile →→ plan → approve → apply → verify

osv.dev
  1. 01

    Lockfile

    Exact installed versions

  2. 02

    Disclosed advisories

  3. 03

    Plan

    Minimum safe bump

  4. 04

    Approve

    TTY, --yes, or --force

  5. 05

    Apply

    Scripts disabled

  6. 06

    Verify

    Rescan + checks

One-sentence flow

Lockfile → → plan → approve (or --force) → rewrite package.json → regenerate lockfile → rescan → run checks → restore both files if anything fails.

1. Discover dependencies

The npm adapter loads package.json and package-lock.json (v2/v3). It inventories exact installed versions, direct vs transitive paths, and development classification.

pnpm and Yarn are detected and rejected clearly. A stale lockfile is an error — run npm install first.

2. Match advisories against OSV

is the vulnerability source of truth. Eridas batches exact (ecosystem, name, version) queries, then loads advisory details: id, aliases, severity, summary, fixed versions, and references.

Unknown-severity advisories stay visible so they are not silently dropped.

3. Plan remediation

For direct dependencies only, Eridas picks the minimum fixed version that closes every advisory for that package, preserves caret/tilde style, and blocks major bumps unless you opt in.

Transitive findings are reported as unresolved instead of injecting risky overrides.

4. Approval gate

eridas fix . --dry-run
eridas fix .
eridas fix . --yes
eridas fix . --force

--force skips the prompt only. It does not skip integrity checks, major policy, rescan, verification, or rollback.

5. Apply and verify

Manifest and lockfile are backed up, updated atomically, regenerated with --ignore-scripts, rescanned against , then checked with configured verification commands. Any failure restores both originals.

Trust boundary

Source of truth

+ lockfile

  • Exact installed versions from package-lock.json
  • Disclosed advisories from osv.dev
  • Fixed versions from advisory range events
  • Rescan after every apply

Not a source of truth

Optional AI

  • Explains confirmed findings only
  • Cannot invent vulnerabilities
  • Cannot choose target versions
  • Soft-fails without changing scan results

Exit codes

CodeMeaning
0Clean scan or successful operation
1Disclosed findings remain at the configured threshold
2Config, lockfile, network, or execution error
3Interactive fix approval declined