Appearance

mk:story-sizer ​

What This Skill Does ​

Sizes a batch of suggested user stories before Jira tickets exist. v1 ships paste-mode only — the user pastes a markdown block of stories, the skill applies deterministic heuristics, and a Story Sizing Report is written to tasks/reports/. Default mode is advisory only. Opt-in --auto-create delegates ticket creation to mk:jira-issue + mk:jira-collaborate behind a single batch-level confirmation gate.

When to Use ​

  • Triggers: "size these stories", "rough-size from spec", "pre-ticket sizing for these user stories", "Fibonacci size this story list".
  • NOT for: single-ticket estimation (mk:jira-estimator) · sprint capacity (mk:planning-engine) · ticket creation alone (mk:jira-issue).

Commands ​

ModeInvocationSide effects
Default/mk:story-sizer --paste [--scout] [--story <id>]Writes a Story Sizing Report. No Jira calls.
Auto-create/mk:story-sizer --paste --auto-create --project <KEY> [--epic <KEY>]After dry-run + single batch confirmation, creates tickets via peer skills.

--scout extracts in-session scout output if present; if not, the skill prompts the user to run /mk:scout first and exits. It never auto-invokes /mk:scout.

Inputs ​

Strict paste-mode markdown template:

text
story: <title>
description: <body — optional>
ac:
  - <ac1>
  - <ac2>
---
story: <next title>
...

Schema, validation rules, and failure modes ([NO_ACS], [MALFORMED_INPUT], length caps, whitespace tolerance) are documented at .claude/skills/story-sizer/references/input-adapter.md. The parser computes a SHA-256 over the verbatim paste body and stamps every record with it — the auto-create source-consistency check relies on this.

Outputs ​

A markdown report at tasks/reports/story-sizing-{YYMMDD}-{slug}.md. Per-story sections include:

  • Title, sized points + complexity, uncertainty band, drivers score breakdown.
  • Description blockquote.
  • Acceptance criteria list.
  • Inconsistencies (when detected).
  • Codebase signals (only when --scout extracted output).
  • Definition of Ready (only when agile-story-gates.md is loaded).
  • Split suggestion (advisory, for 13+ pt or multi-concern stories).
  • REFUSED notice (for stories with no sizeable signal).
  • Suggested mk:jira-issue create command using the v1 field whitelist.

Plus a summary table listing every story at a glance. Idempotency is by source_hash — re-running on an identical paste prompts before overwriting.

Auto-create Safety Gates ​

--auto-create --project <KEY> runs 5 pre-flight checks, in order, ABORTing the batch on any failure:

  1. NO_ACS — any story missing acceptance criteria → ABORT.
  2. Rule-1 injection patterns — Atlassian summary / description fields scanned against the always-loaded inventory at .claude/rules/injection-rules.md (mirrored at references/injection-patterns.md). Any match → ABORT, offending pattern quoted.
  3. Length cap — summary ≤255 chars, description ≤5000 chars.
  4. Duplicate suspect — mk:jira-search JQL project = <KEY> AND summary ~ "<suggested-summary>"; any match aborts the entire batch.
  5. Source-body hash mismatch — recomputes the SHA-256 of the current paste and compares against the report header.

All five pass → markdown dry-run table renders → single AskUserQuestion confirmation → per-ticket two-call sequence:

  • Call A: /mk:jira-issue create --project <KEY> --type Story --summary "..." --story-points <N> --description "..." [--epic <KEY>] [--components <list>] [--labels <list>]
  • Call B: /mk:jira-collaborate add-comment <NEW-KEY> --body "<audit comment>" --internal (team-only; never --public)

Mid-batch Call A failure stops the batch with a manual-cleanup hint (/mk:jira-lifecycle delete <KEY> per stranded ticket). Mid-batch Call B failure logs a WARN row and continues.

v1 Field Whitelist ​

The only flags reaching mk:jira-issue create are: --project, --type, --summary, --story-points, --description, plus optional --epic, --components, --labels. Forbidden in v1 (deferred): assignee, priority, sprint, blocks, and custom-field overrides.

Phase 6 rejects forbidden flags at gating time; Phase 7 re-strips them at call build time as defense in depth.

Skill Rule of Two Compliance ​

Per .claude/rules/injection-rules.md Rule 11:

  • Default mode: 1-of-3 — story-sizer reads untrusted paste content and writes a local report. No sensitive data; no Jira state change.
  • Auto-create mode: 2-of-3 — story-sizer still does not handle credentials. Ticket creation is delegated to peer skills (mk:jira-issue, mk:jira-collaborate) which own the credentialed wrapper. Story-sizer never invokes jira-as directly.

Never 3-of-3.

Audit Comment Customization ​

Default comment body:

text
Initial sizing from mk:story-sizer: {{points}} points (heuristic).
Source: {{report_path}} §{{story_id}}.
Pending team refinement via mk:jira-estimator.

Override path via MEOWKIT_STORY_SIZER_COMMENT_TEMPLATE env var (must contain the three placeholders above). Legacy MEOW_* prefix is NOT honored.

Gotchas ​

  • The source_hash is SHA-256 of the exact paste body. Editing the paste locally between sizing and --auto-create triggers a source-mismatch abort — re-paste and re-run --paste.
  • --scout does not auto-invoke /mk:scout. Run scout first when you want codebase signals.
  • DoR advisory only renders when .claude/rules-conditional/agile-story-gates.md is loaded for the session.
  • No --from-spec / --from-intake adapter in v1. Use paste mode (copy the suggested-user-stories table out of mk:confluence-spec-analyst output and reformat to the paste template).
  • Blocks relationships are not propagated in v1. Link dependencies manually via mk:jira-relationships after creation.
  • Duplicate detection uses Jira's ~ token matching — false positives are possible. Prune or close the matching ticket, then re-run.

See Also ​

Released under the MIT License.

Built with MeowKit