Security model
Keep the project safe no matter who is looking. OSV and the lockfile decide what is vulnerable. Approval and verification decide what changes — with a guided walkthrough or a dense researcher view. Operators: start with the security team runbook.
How to operate Eridas on a team: For security and cyber teams.
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
Sources of truth
| Concern | Authority |
|---|---|
| What is installed | Exact versions in package-lock.json |
| What is disclosed | OSV machine-readable advisories (osv.dev) |
| What to install | Deterministic planner from fixed versions |
| Whether to mutate | Human approval, --yes, or --force |
| Whether the fix worked | OSV rescan + verification commands |
Safe defaults
- Plan before mutate; dry-run available
- Non-interactive sessions cannot mutate without --yes / --force
- Major upgrades require an explicit policy flag
- Transitive issues are reported, not overridden
- npm install during apply uses --ignore-scripts
- Manifest and lockfile restore together on failure
- Scheduled fixes land on a dedicated branch + PR
- Scan-mode workflows request contents: read only
- Advisory text is treated as untrusted before optional AI prompts
Deliberately out of scope for v0.1
- Forum scraping or untrusted “leak” sites
- Auto-merge to the default branch
- Injecting npm overrides for transitive packages
- Letting AI choose fixed versions
- Storing credentials in eridas.config.json