Step 6 — Refine and estimate
mk:jira-evaluator grades ACs, complexity and contradictions; mk:jira-estimator chains off it to suggest story points with an uncertainty annotation. Both are read-only.
Prompt
/mk:jira-evaluator AUTH-201
/mk:jira-estimator AUTH-201What happens behind the scenes
mk:jira-evaluator (read-only):
| Check | Output |
|---|---|
| AC presence | present / missing / vague per AC |
| Complexity dimensions | Code volume, integration count, novelty, risk |
| Inconsistencies | Description ↔ AC contradictions, AC ↔ AC contradictions |
| Verdict | simple / standard / complex |
mk:jira-estimator (read-only, chains off evaluator):
| Input | Output |
|---|---|
| Evaluator complexity + inconsistencies | Suggested story points (1, 2, 3, 5, 8, 13) |
| Uncertainty annotation | If evaluator flagged inconsistencies, estimator says e.g. "5 ± 3 — AC2 wording is ambiguous; estimate halves if redirect target is fixed" |
Estimates are signals, not commitments
Both skills emit suggestions only. The team still does planning poker. The AI provides numbers + reasoning; the humans negotiate.
Sample output
AUTH-201 evaluator
ACs: 3 present, 0 missing, 1 vague (AC2 "lands on /dashboard" — which dashboard?)
Complexity: standard
Inconsistencies: none
Verdict: simple
AUTH-201 estimator
Suggested: 3 points
Uncertainty: ±1 (AC2 wording)
Drivers: 1 new OAuth provider, 1 callback route, session middleware untouchedPrevious: Step 5 — Group into an epic and rank · Next: Step 7 — Tech review against your codebase · All steps
Step 5 — Group into an epic and rank
Epic-add, backlog ranking, and blocks links. Every ordering decision here is dev-declared — the skill never reorders on its own judgment.
Step 7 — Tech review against your codebase
Now that tickets have keys, mk:planning-engine runs per-ticket review and sprint-level planning, and --spec closes the loop back to the original Confluence analysis.