Using Todo Skill: Durable Deferred Work for Coding Agents

Where this fits

You're mid-refactor when your agent flags an unrelated bug it noticed in a dependency. You tell it to "remember that for later," but the next session starts fresh and the note is gone. The using-todo skill solves this by giving agents a durable, cross-session todo store — so deferred work like "audit the auth middleware" or "follow up on the flaky test" survives context resets and actually gets surfaced at the right time.

Why agents benefit

• Persistent memory beyond context windows: Todos live in LWC, not in the conversation, so an agent can pick up where it left off days later.
• Time-aware reminders: Set --target-at timestamps and the agent will see overdue todos as cues in its lifecycle hook, prompting it to act without being told twice.
• Structured parent-child organization: Link related tasks with --parent to keep work grouped without accidentally turning them into plan steps or creating false dependencies.
• Explicit opt-in safety: The skill refuses to run when todo.setting is disabled, preventing agents from silently mutating state the user didn't authorize.
• Searchable and filterable: lwc todo list and lwc todo search let agents quickly rediscover tasks by tag, parent, or keyword instead of guessing from memory.

Practical scenarios

Post-review cleanup backlog

After a code review, your agent notes three refactoring suggestions the user deferred. It adds them as todos with tags like refactor and tech-debt. Two days later, a new session's lifecycle hook surfaces the oldest overdue one, and the agent proactively proposes tackling it.

Multi-session feature work

You're building a feature across several sessions. The agent tracks unfinished subtasks ("write migration script," "update API docs") as todos with parent relationships, so each new session starts with a clear picture of what's outstanding — without re-reading the entire conversation history.

Interrupted research tasks

Your agent was investigating a library upgrade when you switched topics. It saves "evaluate v2 migration path" with a cue describing the key findings so far. When you return, the reminder includes enough context to resume without re-doing the research.

Add it to your agent workflow

Install the skill and enable the todo setting:

npx skills add JanYork/llm-wiki-cli --skill using-todo
lwc config set --todo enabled

Then have your agent capture deferred work:

lwc todo add "Audit auth middleware" --tag security --cue "Found potential timing attack" --target-at 2025-06-15T10:00:00Z

The agent will now surface this todo when its target time arrives, keeping your deferred work alive across sessions.

Compare before adopting

Before committing, compare against alternatives: check the repo's commit freshness and issue responsiveness on GitHub, evaluate whether LWC's storage model fits your agent's runtime, and weigh it against simpler approaches like appending to a markdown file. The skill's explicit opt-in and safety guardrails are strong signals, but verify they match your workflow.

Why it is worth tracking

With 44 stars and an active maintainer, this skill is early but well-scoped. Its clean separation between todos and plans, plus the opt-in consent model, shows thoughtful design. Watch for growth in adoption and community feedback — if it gains traction, it could become a standard building block for persistent agent memory. Evaluate it now if you're already using LWC or struggling with cross-session context loss. See how it compares to other agent skills at OpenAgentSkill.