/pm-autoscope — Automated Scoping

Automates the discovery and creation pipeline for epics, stories, and tasks.

How it works

1. •Call pm_auto_scope() to detect the project state and get discovery signals.
2. •Based on the mode returned, follow the appropriate workflow below.

Full Scan Mode (new projects)

Triggered when no epics or stories exist yet. The tool returns codebase signals (docs, build files, source tree).

Workflow:

1. •Analyze the codebase signals returned by pm_auto_scope
2. •Propose 2-5 epics covering the major areas of work
3. •Present the epic list to the user for approval — let them add/remove/edit
4. •Create approved epics with pm_create_epic
5. •For each epic, propose 2-6 stories
6. •Present stories to the user for approval per epic
7. •Create approved stories with pm_create_story, linking to the epic via epic_id
8. •For each story, propose 2-6 tasks
9. •Present tasks to the user for quick approval
10. •Create approved tasks with pm_create_task
11. •Show summary: N epics, N stories, N tasks created, total points

Incremental Mode (existing stories need tasks)

Triggered when stories exist but some have no tasks. Results are paginated — the tool returns a batch of stories at a time (default 5).

Workflow:

1. •Call pm_auto_scope() — returns the first batch of undecomposed stories (IDs and titles only, no bodies)
2. •For each story in the batch: a. Call pm_scope(story_id) to get full decomposition context (story body + existing tasks + guidance) b. Propose 2-6 tasks for the story based on its content c. Present the task list to the user for quick approval (yes/edit/skip) d. Create approved tasks with pm_create_tasks(story_id, tasks) (batch — one call per story)
3. •If has_more is true in the response, call pm_auto_scope(offset=next_offset) for the next batch
4. •Repeat steps 2-3 until all stories are scoped (has_more is false)
5. •After all batches are done, show summary:
   • •N stories scoped
   • •N tasks created
   • •Total points assigned
6. •Suggest next steps: /pm board to see available work or /pm-plan for sprint planning

Guidelines

• •Keep task titles as verb phrases: "Add authentication middleware", "Write unit tests for parser"
• •Each task should be 1-5 points (completable in one session)
• •Tasks should be independently testable
• •First task in a story sets up the foundation
• •Last task handles integration/cleanup
• •Don't over-decompose — 2-6 tasks per story is the sweet spot

Forcing a mode

Users can force a specific mode:

• •/pm autoscope full — force full codebase scan even if stories exist
• •/pm autoscope incremental — force incremental even if no stories exist