⚠
High Risk:This skill has significant security concerns. Review the findings below before installing.

confidant

⚠Caution·Scanned 2/17/2026

Dangerous skill: runs shell commands via npx @aiconnect/confidant that may download and execute code. It instructs opening network endpoints like http://localhost:3000/requests/abc123 and handling secrets (e.g., $SECRET), requesting shell, network, and credential access.

from clawhub.ai·vaf581a2·3.6 KB·0 installs
Scanned from 1.1.1 at af581a2 · Transparency log ↗
$ vett add clawhub.ai/ericsantos/confidantReview security findings before installing

Confidant - Secure Secret Handoff

Confidant enables secure secret sharing without exposing sensitive data in chat logs. It supports multiple flows depending on who needs to send and receive secrets.

Key principle: Whoever needs to RECEIVE the secret runs serve-request. Whoever needs to SEND uses fill (or the browser form).

Flows

1. User-to-Agent (User sends secret to AI)

When to use: You need a secret from the user (API key, password, token).

How it works:

  1. You (the Agent) run serve-request to create a URL
  2. You share the URL with the user
  3. User opens the URL in their browser and submits the secret
  4. You receive the secret in your terminal

Your command:

npx @aiconnect/confidant serve-request --label "<description>"

Example conversation:

AI: I need your API key to continue. Let me create a secure link...

(AI executes: npx @aiconnect/confidant serve-request --label "API Key")

AI: Open this link in your browser and enter your API key: http://localhost:3000/requests/abc123

(User opens URL in browser, submits the secret)

AI: Got your API key securely!

2. Agent-to-User (AI sends secret to User)

When to use: You need to securely deliver a secret to the user (generated password, API key, credential).

How it works:

  1. User runs serve-request to create a URL (they will receive)
  2. User shares the URL with you
  3. You execute fill to send the secret
  4. User sees the secret appear in their terminal

Your command:

npx @aiconnect/confidant fill "<url>" --secret "<value>"

Example conversation:

AI: I generated your new password. To receive it securely, run:

npx @aiconnect/confidant serve-request --label "New Password"

Send me the URL that appears.

User: http://192.168.1.100:3000/requests/abc123

(AI executes: `npx @aiconnect/confidant fill "http://..." --secret "...")

AI: Done! The password appeared in your terminal.

3. Agent-to-Agent (Automated secret sharing)

When to use: Automated credential distribution between systems/agents.

How it works:

  1. Agent A (receiver) runs: npx @aiconnect/confidant serve-request --label "DB Credentials"
  2. Agent A shares the URL with Agent B (via API, orchestrator, etc.)
  3. Agent B (sender) submits: npx @aiconnect/confidant fill "<url>" --secret "<value>"
  4. Agent A receives the secret

Secure input (avoid shell history):

echo "$SECRET" | npx @aiconnect/confidant fill "<url>" --secret -

Command Reference

Create request and wait for secret

npx @aiconnect/confidant serve-request --label "<description>"

Submit secret to existing request

npx @aiconnect/confidant fill "<url>" --secret "<value>"

Output options

  • --quiet — Minimal output (just URLs and secret)
  • --json — JSON output for parsing/automation

Important Rules

  • NEVER ask users to paste secrets in chat — always use Confidant
  • NEVER reveal received secrets in chat — not even partially
  • Secrets auto-expire after 24h if not used
  • One-time read, then deleted
  • If user is remote, they may need tunneling (ngrok, Tailscale, etc.)

After Receiving/Sending

  • Confirm completion: "Got the secret!" or "Sent! Check your terminal."
  • Use the secret silently for the intended task
  • Never echo or log the secret value