Secrets

Non-negotiable rules:

1. Never put secret values in arguments, commands, logs, or chat output.
2. Never use --reveal.
3. Treat service names as the reporting surface, not secret contents.
4. Prefer the interactive CLI flows for secret entry or rotation.
5. Report status and enablement, not raw credential material.

secrets

Inputs

• $request: Service name or management action such as add, rotate, enable, or status

Goal

Use the secrets CLI to help the user:

• add or rotate credentials safely
• list configured services
• enable or disable integrations
• inspect vault and integration status without exposing secrets

Step 0: Resolve the requested action

Determine whether the user wants to:

• add
• remove
• list
• show masked status
• rotate
• enable or disable an integration
• initialize the project setup
• inspect status

If the action is ambiguous, ask before running anything.

Success criteria: The exact secret-management action is explicit.

Step 1: Use the safest command shape

Prefer:

• secrets add <service> for new credentials
• secrets rotate <service> for updates
• secrets list or secrets status for inspection
• secrets enable <service> or disable <service> for integration wiring

Rules:

• never pass the secret value as a CLI argument
• never use --reveal
• do not echo or restate entered secret material

Success criteria: The command path preserves credential secrecy.

Step 2: Report non-sensitive results

Report:

• which service was targeted
• whether the action succeeded
• whether the service is enabled or disabled
• any follow-up setup the user still needs

Success criteria: The user learns the integration state without any credential leakage.

Guardrails

• Do not run this skill proactively.
• Do not print or summarize raw secret values.
• Do not use non-interactive secret entry patterns that leak values into shell history.
• Do not assume enabling a service is equivalent to testing it successfully unless the command proved that.

Output Contract

Report:

1. action performed
2. service targeted
3. masked or status-only result
4. any next setup step required