Native autonomous agent integrations

[rodion-m]

Goal

Extend Agentic Secrets from registered local CLI delivery into native workflows for OpenClaw, Hermes Agent, and other autonomous agents without weakening the local secret-authority boundary.

Scope

• Design agent-facing integration patterns that expose bounded capabilities, pinned profiles, and redacted audit events.
• Avoid raw secret retrieval APIs and keep secret delivery explicit, narrow, approved, auditable, and fail-closed.
• Study approaches, implementations, and product ideas from https://github.com/The-17/agentsecrets while preserving Agentic Secrets stricter local approval and secret-delivery boundaries.

Milestones

1. Inventory target autonomous agent workflows, starting with OpenClaw and Hermes Agent.
2. Define native integration contracts for bounded capabilities and pinned profiles.
3. Add redacted audit events for agent-native workflows.
4. Prototype one integration without exposing raw secret retrieval.
5. Document which AgentSecrets ideas were adopted, rejected, or adapted.

Acceptance criteria

• Integrations do not introduce public plaintext secret retrieval APIs.
• Agent actions use pinned profiles or equivalent bounded delivery contracts.
• Audit records identify the agent workflow and delivered secret alias without secret values.
• Borrowed patterns from The-17/agentsecrets are reviewed against Agentic Secrets approval, secret-authority, and fail-closed guarantees.

Roadmap source: ROADMAP.md, Direction 3.