/plan-tasks

Plan Tasks

Transform conversation planning and requirements into a markdown task list where each task is completely self-contained with all necessary context inline.

Task Format

The task list is in .llm/todo.md.

NEVER use the Read tool on .llm/todo.md. Always interact with the task list exclusively through the Python scripts.

Task States

• [ ] - Not started (ready to work on)
• [x] - Completed
• [!] - Blocked after failed attempt

Standalone Context

Each task is extracted and executed in isolation. The task_get.py script extracts only one task at a time - it cannot see other tasks in the file. Therefore:

1. Every task must contain ALL context needed to implement it
2. Repeat shared context in every related task - if 5 tasks share the same background, repeat it 5 times
3. Never reference other tasks - phrases like "similar to task above" are useless
4. Include the full picture - source of inspiration, files involved, patterns to follow
5. If a plan file exists (e.g., ~/.claude/plans/*.md), archive it locally before adding any tasks:
   • Create the .llm/plans/ directory if it does not exist
   • Move the file from ~/.claude/plans/<auto-generated-name>.md to .llm/plans/<yyyy-mm-dd>-<descriptive-name>.md
   • Choose a concise, meaningful name (e.g., 2025-12-04-thread-safety-tests.md)
   • Include the archived .llm/plans/ absolute path in each task

When to Use

Use this command at the end of a planning conversation when you have discussed requirements, approaches, and implementation details but have not started coding yet. This captures the conversation context into actionable tasks in .llm/todo.md.

Input

The input is the current conversation where planning and requirements have been discussed. Transform the plans, ideas, and requirements from the discussion into self-contained tasks in a markdown checklist format, appended to .llm/todo.md.

Add every task in a single bash command that chains one task_add.py call per task with &&:

python ${CLAUDE_PLUGIN_ROOT}/scripts/task_add.py .llm/todo.md "First task
  Context line 1" && \
python ${CLAUDE_PLUGIN_ROOT}/scripts/task_add.py .llm/todo.md "Second task" && \
python ${CLAUDE_PLUGIN_ROOT}/scripts/task_add.py .llm/todo.md "Third task"

Running the whole batch as one command keeps the write window to .llm/todo.md extremely short, so concurrent sessions writing to the same file are far less likely to interleave their tasks. Never add tasks across separate commands.

Task Writing Guidelines

Each task should be written so it can be read independently from - [ ] to the next - [ ] and contain:

1. Full absolute paths - Never use relative paths
2. Exact class/function names - Specify exact names of code elements
3. Analogies to existing code - Reference similar existing implementations
4. Specific implementation details - List concrete methods or operations
5. Module/package context - State which module or package the work belongs to
6. Dependencies and prerequisites - Note what needs to exist or be imported
7. Expected outcomes - Describe what success looks like

Example

- [ ] Create a new test class `SynchronizedBagTest` at `/Users/craig/projects/eclipse-collections/unit-tests-thread-safety/src/test/java/org/eclipse/collections/impl/bag/mutable/SynchronizedBagTest.java` to test thread-safety of `org.eclipse.collections.impl.bag.mutable.SynchronizedBag`. Similar to how `SynchronizedMutableListTest` covers `SynchronizedMutableList`, this should extend `SynchronizedTestTrait` and implement test traits like `SynchronizedCollectionTestTrait`, `SynchronizedMutableIterableTestTrait`, and `SynchronizedRichIterableTestTrait`. The test should verify that all public methods of SynchronizedBag properly synchronize on the lock object using the `assertSynchronized()` method. Include tests for bag-specific methods like `addOccurrences()`, `removeOccurrences()`, `occurrencesOf()`, `forEachWithOccurrences()`, and `toMapOfItemToCount()`.