story-sizer

The story-sizer agent applies deterministic heuristics to a batch of suggested user stories and emits a Story Sizing Report with Fibonacci point estimates, complexity verdicts, inconsistency flags, and optional split proposals. It never mutates Jira directly — auto-create mode delegates to mk:jira-issue and mk:jira-collaborate.

Cognitive Framing

"Size the work before the ticket exists, so the ticket creation step is informed instead of arbitrary."

The story-sizer is a read-only sizing agent. It does NOT write story points to existing Jira tickets (jira-estimator does that for single tickets). It produces an advisory artifact for the dev to review, and — when the user opts in — drives a single-confirmation batch handoff to peer execution skills.

Key Facts

Type	Intelligence (pre-ticket)
Phase	On-demand
Model	inherit
Color	orange
Safety	Read-only at Jira. Local-FS write to tasks/reports/. Skill Rule of Two: 1-of-3 default · 2-of-3 with --auto-create (delegated; never 3-of-3).
Never does	Read Jira credentials · invoke the credentialed Jira wrapper · call Atlassian REST directly · auto-invoke /mk:scout · use --public for audit comments · emit v1-forbidden flags (assignee, priority, sprint, blocks, custom-fields).

When to Use

• When you need rough sizing for a batch of stories before tickets exist.
• After a mk:confluence-spec-analyst run that produced suggested user stories — reformat to the paste template and run story-sizer.
• When you want to delegate ticket creation behind a single confirmation gate instead of running mk:jira-issue create once per story.

NOT For

• Sizing a single existing Jira ticket — use jira-estimator.
• Sprint capacity analysis — use mk:planning-engine.
• Auto-invoking scout, plan-creator, or any other skill.

Sizing Flow

1. Receive List<StoryRecord> parsed by scripts/parse-paste-stories.py.
2. Apply heuristics from references/sizing-heuristics.md (5 dimensions, integer math).
3. Emit per-story points, uncertainty, complexity, inconsistencies, optional split_proposal, optional codebase_signals, optional dor_status.
4. Refuse to size stories with [NO_ACS] flag from the adapter.
5. Render via the template + writer (scripts/write-sizing-report.py).
6. Return the report path.

Auto-create Handoff

The agent does not call Jira itself. The SKILL.md orchestration layer:

1. Runs the 5 pre-flight checks (references/auto-create-gating.md).
2. Renders the dry-run table.
3. Asks the user one yes/no question.
4. Executes the per-ticket two-call sequence (references/auto-create-execution.md):
   • Call A: /mk:jira-issue create ...
   • Call B: /mk:jira-collaborate add-comment ... --internal
5. Appends ## Created Tickets to the report.

Determinism

Same StoryRecord produces the same points / drivers_score / complexity every run. Reasoning prose may vary; the numeric output does not.

Gotchas

• Source-body hash records the SHA-256 of the paste; editing locally between sizing and auto-create triggers a mismatch ABORT.
• DoR advisory only renders when Agile rules are loaded for the session.
• Comment template can be overridden via the MEOWKIT_STORY_SIZER_COMMENT_TEMPLATE env var (file path; must include the three placeholders).

See Also

• Skill: mk:story-sizer
• Peer execution: jira-issue, jira-collaborate
• Peer intelligence: jira-evaluator, jira-estimator, jira-analyst
• Workflow: Spec → PR walkthrough — Step 3.5