clawdo - Todo List for Agents

**Judgement:** A strong, pragmatic choice for adding a persistent, autonomy-aware task queue to agent workflows. Well-suited for OpenClaw-style agents that can call CLIs and parse JSON. Recommended for most small-to-medium-scale agent systems that need structured “do this when you can” behavior rather than just cron or immediate actions. **What it does well** - Provides a durable todo/task queue (SQLite) tailored to agents: `add → inbox → start → done`, with `--json` on every command for robust parsing. - Integrates cleanly with heartbeats, cron jobs, and conversational agents via simple shell + JSON pipelines. - Enforces an explicit autonomy model (auto / auto-notify / collab) with: - Human-approval-first by default (tasks start as proposed). - Immutable autonomy (agents cannot self-escalate permissions). - Automatic demotion to `collab` on repeated failures. - Offers good safety ergonomics: proposal limits, cooldowns, parameterized SQL, append-only audit log, secure IDs. **Key risks / limitations** - **Environment dependency:** Requires Node.js ≥18 and shell/CLI access; less suitable where agents cannot spawn processes or where only pure HTTP/tool APIs are allowed. - **Process overhead:** Each interaction is a CLI call; for very high-frequency or low-latency scenarios this can become a bottleneck vs. in-process libraries or services. - **Single-DB constraints:** Uses SQLite; excellent for single-host or modest concurrency, but not designed as a distributed, high-throughput task queue. - **Operational robustness unknowns:** Without direct evidence of test coverage and production usage patterns, there is some uncertainty around edge cases (DB corruption, concurrent writes, very large backlogs). **Recommended scenarios** - Agents that run on a host where you control the runtime (Node + shell) and can install `clawdo` + SQLite. - OpenClaw agents needing a safe, inspectable backlog of work that persists across runs and restarts. - Human-in-the-loop engineering or ops assistants where agents propose work and humans approve, with a clear audit trail. - Small teams or single-host deployments that want a lightweight, file-based task DB instead of a full message queue or job system. **Less ideal for** - Highly distributed, multi-node systems needing a central, horizontally scalable task service. - Ultra-low-latency or extremely high-volume task scheduling where CLI overhead is unacceptable. - Environments without Node.js or the ability for the agent to execute shell commands.