For agents
Eridas isn’t an “agent product” — it’s a deterministic dependency-security service that’s easy to set up, read, use, and debug. Coding agents like it for the same reason CI and humans do.
Agent skill
Drop Eridas into any coding agent
One portable SKILL.md so Cursor, Claude Code, Codex, Copilot, and others can use the same deterministic CLI and API humans do. Advisory truth from OSV— dry-run before mutate.
/skills/eridas/SKILL.md
Install paths
Drop the skill into your project (or global skills folder). Cursor and many others read .agents/skills/ automatically.
mkdir -p .agents/skills/eridas && curl -fsSL https://erydas.vercel.app/skills/eridas/SKILL.md -o .agents/skills/eridas/SKILL.md| Agent | Project path |
|---|---|
| Any / Cursor | .agents/skills/eridas/SKILL.md |
| Cursor (alt) | .cursor/skills/eridas/SKILL.md |
| Claude Code | .claude/skills/eridas/SKILL.md |
| Codex | .codex/skills/eridas/SKILL.md |
Same file everywhere — only the folder name changes. Personal/global installs use ~/.agents/skills/eridas/ (or your agent’s equivalent).
Read this first
- Vulnerability truth comes from
. Do not invent advisories. - Prefer flags over prompts. Use --dry-run before mutating.
- --force skips approval prompts only — not integrity checks or major-upgrade policy.
- Optional --ai explains confirmed findings; it must not choose versions.
Machine indexes: /agents.txt · /llms.txt · Skill: /skills/eridas/SKILL.md
Safe agent workflow
# 1. Discover disclosed issues (read-only)
npx eridas scan . --quiet --format json
# 2. Preview remediation without writing files
npx eridas fix . --dry-run --quiet --format json
# 3. Apply only when the environment is allowed to mutate
npx eridas fix . --yes
# 4. Optional: schedule PR-based remediation
npx eridas init . --mode fixEridas Cloud + GitHub
When the org has the GitHub App connected, agents can drive remote repos with an API key (erd_…) via CLI, @eridas/cloud-sdk, or MCP.
export ERIDAS_API_KEY=erd_…
export ERIDAS_API_URL=https://erydas.vercel.app
eridas auth whoami --format json
eridas cloud repos --format json
eridas cloud scan --repo-id <uuid> --wait --format json
eridas cloud findings --repo-id <uuid> --format json
# Prefer dry-run locally; only remediate with explicit user intent:
eridas cloud remediate --repo-id <uuid> --wait --format jsonSee Docs for /api/v1/me, sync, git-status, package-root, and ?wait=true long-poll.
MCP (Cursor and other editors)
Install @eridas/mcp and point your editor at eridas-mcp over stdio.
{
"mcpServers": {
"eridas": {
"command": "npx",
"args": ["-y", "@eridas/mcp"],
"env": {
"ERIDAS_API_KEY": "erd_…",
"ERIDAS_API_URL": "https://erydas.vercel.app"
}
}
}
}Local tools always work. Cloud tools need ERIDAS_API_KEY. Remediation requires confirm=true.
Exit codes
| Code | Meaning | Agent action |
|---|---|---|
| 0 | Clean / success | Continue |
| 1 | Findings remain | Report findings; do not claim the tree is clean |
| 2 | Operational error | Fix lockfile/config/network; retry |
| 3 | Approval declined | Do not retry apply without explicit user intent |