[Phase 1] Bake the fixed critic rubric into Bob's planning prompt

[github-actions]

Story

As a maintainer relying on Bob's plans,
I want the BMAD Scrum Master prompt to carry the fixed quality rubric as an explicit self-check before it emits plan.json,
so that even before a separate critic step exists, Bob screens its own plan against the rubric, capturing much of the critic's value at near-zero cost.

Acceptance Criteria

1. prompts/bmad/scrum-master.md gains a named quality-rubric section listing the fixed checks from discussion 💡 Adversarial Plan-Critic + Structural Gates for the Initiative-Planner (Bob) #593: no contested open-question baked into an acceptance criterion; an initiative-level success metric is present; a cost cap is present; untracked prerequisites are surfaced; each story is independently reviewable; and for eval/optimization stories, overfitting/reward-hacking and artifact-immutability are addressed.
2. Each rubric item is phrased as a yes/no self-check, and the rubric instructs Bob to run it as a FINAL pass and route any unresolved item to open_questions rather than guessing (consistent with the existing no-invented-AC hard rule).
3. The rubric lives in exactly one place (scrum-master.md, which the workflow prompt already instructs Bob to cat); the workflow inline prompt gains only a one-line pointer to run the rubric self-check, with no divergent second copy of the checklist.
4. No schema or script behavior changes; the existing test suite still passes.

Tasks / Subtasks

• Add the quality-rubric section to prompts/bmad/scrum-master.md as a fixed, quotable checklist of the six checks. (AC: test issue from agent #1, Go-live improvements for PR review agent #2)
• Add a one-line instruction to the workflow inline prompt to run the rubric self-check before emitting the plan, pointing at scrum-master.md (single source of truth). (AC: Add @claude delegation, auto-merge, and rebase handling #3)
• Verify no schema/script changes are needed and existing bats pass. (AC: Optimize review: small-PR and incremental fast paths #4)

Dev Notes

• prompts/bmad/scrum-master.md is the single source Bob reads — the workflow inline prompt at .github/workflows/initiative-planner.yml:117-119 already tells Bob to cat it. Put the rubric there; do NOT fork a divergent copy into the workflow YAML.
• The six findings the hand review of Initiative: Eval-gated, human-reviewed self-improving skills (SkillOpt-style) #581 surfaced (per discussion 💡 Adversarial Plan-Critic + Structural Gates for the Initiative-Planner (Bob) #593) ARE the rubric: contested-AC baked into ACs, no initiative success metric, no cost cap, reward-hacking/overfitting, untracked prereqs, and story reviewability.
• scrum-master.md already has a hard rule (No invented acceptance criteria / unresolved -> open_questions) — extend it, do not contradict it.
• This story defines the CANONICAL rubric text that the Phase-2 critic story (id 6) reuses verbatim — keep it self-contained and quotable so a second step can load it.

Project Structure Notes

Prompt-only change. Keep the rubric in scrum-master.md as the single source; the workflow already sources it, avoiding the drift that two copies would cause.

References

• prompts/bmad/scrum-master.md
• .github/workflows/initiative-planner.yml#L103-L176
• tests/test_initiative_planner.bats

Likely target surface

• prompts/bmad/scrum-master.md
• .github/workflows/initiative-planner.yml

Story prepared by the BMAD Scrum Master (Bob) for epic #597. Status: ready-for-dev.

[don-petry]

Dev-Lead Implementation Plan

Issue: #599 — [Phase 1] Bake the fixed critic rubric into Bob's planning prompt

Scope

Add a named, quotable six-item quality-rubric self-check to prompts/bmad/scrum-master.md (single source of truth) and a one-line pointer to it in the initiative-planner workflow's inline prompt — no schema/script changes.

Implementation Checklist

• Add a Quality rubric section to prompts/bmad/scrum-master.md: six fixed yes/no checks (contested-AC, initiative success metric, cost cap, untracked prereqs, story reviewability, eval/optimization overfitting+artifact-immutability), run as a FINAL pass, unresolved → open_questions, consistent with the existing no-invented-AC hard rule.
• Add a one-line pointer in .github/workflows/initiative-planner.yml inline prompt instructing Bob to run the rubric self-check before emitting the plan (no divergent second copy of the checklist).

Tests to Write

• Pure documentation/prompt change with no observable script/schema behavior — Phase 2 (TDD) exception applies. No new tests; the existing bats suite must still pass unchanged (AC Optimize review: small-PR and incremental fast paths #4).

Test Command

bats tests/test_initiative_planner.bats (plus full bats tests/ and bash .dev-lead/scripts/dev-lead-lint.sh)

[don-petry]

Dev-Lead: Implementation Complete

Plan Execution

• Added a named Quality rubric section to prompts/bmad/scrum-master.md — six fixed yes/no self-checks (contested-AC, initiative success metric, cost cap, untracked prerequisites, story reviewability, eval/optimization overfitting+artifact-immutability), run as a FINAL pass, unresolved items routed to open_questions, explicitly consistent with the existing no-invented-AC hard rule.
• Added a one-line pointer in .github/workflows/initiative-planner.yml inline prompt to run the rubric self-check before emitting the plan — no divergent second copy of the checklist (single source of truth in scrum-master.md).
• Verified no schema/script changes were needed; existing bats suite still passes.

AC coverage

• AC#1 — six fixed checks from discussion 💡 Adversarial Plan-Critic + Structural Gates for the Initiative-Planner (Bob) #593 present as a named rubric section. ✅
• AC#2 — each item is a yes/no self-check; instructed to run as a FINAL pass and route unresolved items to open_questions rather than guess. ✅
• AC#3 — rubric lives only in scrum-master.md; workflow gains a single pointer, no duplicated checklist. ✅
• AC#4 — no schema/script behavior changed; suite passes. ✅

Phase 2 (TDD) note

This is a pure prompt/documentation change with no observable script or schema behavior change, so the Phase 2 test-first exception applies — no new tests were added. The existing suite is the regression guard.

Test Results

# tests/test_initiative_planner.bats (the suite covering the touched surface)
1..11
ok 1 validate-plan accepts a well-formed acyclic plan
ok 2 validate-plan rejects a self-blocking story
ok 3 validate-plan rejects a dependency cycle (with an entry point present)
ok 4 validate-plan rejects a dangling blocked_by reference
ok 5 validate-plan rejects a plan with no entry point
ok 6 apply-plan (DRY_RUN) creates epic + stories + edges and posts comment
ok 7 apply-plan threads distinct issue numbers in dry-run
ok 8 apply-plan NEVER applies initiative:auto to any created issue
ok 9 apply-plan marks hands_off story with hold + hands-off labels
ok 10 existing-issue prerequisite edge references the real issue number
ok 11 story bodies are rendered in the BMAD create-story template

# repo lint: bash .dev-lead/scripts/dev-lead-lint.sh
  [lint] shellcheck on 45 file(s)...
  [lint] validating 4 agent profile(s)...
  [lint] all checks passed

# workflow YAML parses OK

Files Changed

• prompts/bmad/scrum-master.md: new "Quality rubric — final self-check before emitting the plan" section (canonical, quotable for the Phase-2 critic story) + a one-line forward pointer in Step 5.
• .github/workflows/initiative-planner.yml: one-line pointer in the inline prompt to run the rubric self-check before emitting the plan.

Notes

• Pre-existing, unrelated test failure: tests/test_batch_fallback.bats → "batch: Gemini trust error (55) falls back to Copilot" fails on a clean tree as well (confirmed via git stash), stemming from the recent fallback-reorder commit afb4acd — not introduced by this prompt-only change and outside its scope.
• bats is not installed locally; tests were run via npx bats.