Agent Coordination with dex
Command Invocation
Use dex directly for all commands. If not on PATH, use npx @zeeg/dex instead.
command -v dex &>/dev/null && echo "use: dex" || echo "use: npx @zeeg/dex"Core Principle: Tickets, Not Todos
Dex tasks are tickets - structured artifacts with comprehensive context:
- Name: One-line summary (issue title)
- Description: Full background, requirements, approach (issue body)
- Result: Implementation details, decisions, outcomes (PR description)
Think: "Would someone understand the what, why, and how from this task alone?"
Dex Tasks are Ephemeral
Never reference dex task IDs in external artifacts (commits, PRs, docs). Task IDs like abc123 become meaningless once tasks are completed. Describe the work itself, not the task that tracked it.
When to Use dex
Use dex when:
- Breaking down complexity into subtasks
- Work spans multiple sessions
- Context needs to persist for handoffs
- Recording decisions for future reference
Skip dex when:
- Work is a single atomic action
- Everything fits in one session with no follow-up
- Overhead exceeds value
dex vs Built-in Task Tools
Some AI agents (like Claude Code) have built-in task tools. These are session-only and not the same as dex.
| dex | Built-in Task Tools | |
|---|---|---|
| Persistence | Files in .dex/ | Session-only |
| Context | Rich (description + context + result) | Basic |
| Hierarchy | 3-level (epic → task → subtask) | Flat |
Use dex for persistent work. Use built-in task tools for ephemeral in-session tracking only.
Basic Workflow
Create a Task
dex create "Short name" --description "Full implementation context"Description should include: what needs to be done, why, implementation approach, and acceptance criteria. See examples.md for good/bad examples.
List and View Tasks
dex list # Pending tasks
dex list --ready # Unblocked tasks
dex show <id> # Full detailsComplete a Task
dex complete <id> --result "What was accomplished" --commit <sha>GitHub/Shortcut-linked tasks require either --commit <sha> or --no-commit:
- Use
--commit <sha>when you have code changes (issue closes when merged) - Use
--no-commitfor non-code tasks like planning or design (issue stays open)
Always verify before completing. Results must include evidence: test counts, build status, manual testing outcomes. See verification.md for the full checklist.
Edit and Delete
dex edit <id> --description "Updated description"
dex delete <id>For full CLI reference including blockers, see cli-reference.md.
Understanding Task Fields
Tasks have two text fields:
- Name: Brief one-line summary (shown in
dex list) - Description: Full details - requirements, approach, acceptance criteria (shown with
--full)
When you run dex show <id>, the description may be truncated. The CLI will hint at --full if there's more content.
Gathering Context
When picking up a task, gather all relevant context:
dex show <id> --full # Full task details
dex show <parent-id> --full # Parent context (if applicable)
dex show <blocker-id> --full # What blockers accomplishedBefore starting, verify you can answer:
- What needs to be done specifically?
- Why is this needed?
- How should it be implemented?
- When is it done (acceptance criteria)?
If any answer is unclear:
- Check parent task or completed blockers for more details
- Suggest entering plan mode to flesh out requirements before starting
Proceed without full context when:
- Task is trivial/atomic (e.g., "Add.gitignore entry")
- Conversation already provides the missing context
- Description itself is sufficiently detailed
Task Hierarchies
Three levels: Epic (large initiative) → Task (significant work) → Subtask (atomic step).
Choosing the right level:
- Small feature (1-2 files) → Single task
- Medium feature (3-7 steps) → Task with subtasks
- Large initiative (5+ tasks) → Epic with tasks
# Create subtask under parent
dex create --parent <id> "Subtask name" --description "..."For detailed hierarchy guidance, see hierarchies.md.
Recording Results
Complete tasks immediately after implementing AND verifying:
- Capture decisions while fresh
- Note deviations from plan
- Document verification performed
- Create follow-up tasks for tech debt
Your result must include explicit verification evidence. Don't just describe what you did—prove it works. See verification.md.
Commit Messages with GitHub Issues
When a task is linked to a GitHub issue (shown in dex show output), include issue references in commit messages:
- Root tasks (the task itself has GitHub metadata): Use
Fixes #N
- This closes the issue when merged
- Subtasks (parent/ancestor has GitHub metadata): Use
Refs #N
- This links to the issue without closing it
Check dex show <id> for GitHub issue info before committing. The "(via parent)" indicator means use Refs, direct metadata means use Fixes.
Best Practices
- Right-size tasks: Completable in one focused session
- Clear completion criteria: Description should define "done"
- Don't over-decompose: 3-7 children per parent
- Action-oriented descriptions: Start with verbs ("Add", "Fix", "Update")
- Verify before completing: Tests passing, manual testing done
Additional Resources
- cli-reference.md - Full CLI documentation
- examples.md - Good/bad context and result examples
- verification.md - Verification checklist and process
- hierarchies.md - Epic/task/subtask organization