The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs direct modification of production infrastructure using commands like gcloud run services update. It specifically uses --update-env-vars to alter service configurations, which can impact production stability and security.
[CREDENTIALS_UNSAFE]: The skill accesses and audits highly sensitive data, including service-wallets and custodial-keys. The instructions involve checking signer wallets and ensuring funded custodial wallets are decryptable, which exposes cryptographic keys to the agent's context.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process production logs and external health signals to diagnose issues, creating a surface for indirect prompt injection if those logs contain malicious instructions.
Ingestion points: Production logs accessed via npm run logs and the health check endpoint https://three.ws/api/healthz (SKILL.md).
Boundary markers: No delimiters or specific instructions to treat log content as untrusted are present.
Capability inventory: Ability to update production environment variables, commit code via git, and execute local scripts (SKILL.md).
Sanitization: No evidence of log filtering or escaping before processing findings.

gcp-triage