Roadrunner

**Judgement:** Roadrunner is a strong, feature-rich skill for automating and querying Beeper Desktop via its local CLI. It’s well-suited for power users who explicitly want an agent to interact with their real chats and messages, but it carries inherent side-effect risks and requires correct local setup. **What it’s good for** - High-fidelity access to Beeper data: listing/searching chats and messages, unread rollups, status, and global search. - Operational workflows: monitoring chats (tail/wait), summarizing conversations, locating attachments, and managing reminders. - Agent-safe integration: `--agent`, `--json`, `--envelope`, `--readonly`, `--enable-commands`, and pagination controls give the agent precise, structured, and bounded access. **Key risks / limitations** - **Side-effectful operations:** The skill can send/edit messages, create/archive chats, set reminders, and upload/send files. Misinterpretation of user intent or ambiguous chat IDs could lead to unintended messages or chat state changes. - **Privacy and data exposure:** While output is local, the agent sees real message content. Summaries or answers must avoid dumping raw JSON or unnecessary sensitive details. - **Environment coupling:** Requires Beeper Desktop running, valid auth, and `rr` installed; unusable outside that environment. Some features (e.g., websocket events) are experimental and may fail or require fallbacks. **Recommended scenarios** - Users who explicitly say they want to control Beeper Desktop or Beeper chats via CLI/automation. - Agents that help triage large volumes of messages (search, filter, summarize) or monitor for specific keywords/events. - Workflow automation where read-only queries are primary, with occasional, clearly requested mutations (e.g., “send this exact message to chat X”, “set a 2h reminder on this thread”). **Use with extra caution when** - The user’s request is vague about recipients or exact message content (the agent should clarify before any mutation). - The user is not clearly aware that the agent is operating on **real** chats; in such cases prefer read-only operations or confirm explicitly before writing. - Handling especially sensitive conversations, where even summarized exposure to the agent may be undesirable.