j-review-plan

name: j-review-plan description: Review plan for TDDAB/Step conformity. Checks methodology compliance, dependency ordering, and validates with CVM parsePlan.

Prerequisites

• Read j-settings.md if it exists (provides methodology config)
• Read (if not already done) MB — skip if memory-bank/ does not exist yet

Steps

1. Read Methodology Reference

Check j-settings.md @backend-method (if j-settings.md exists):

• If tddab → load the /tddab-planner skill (its rules are the source of truth). If @tddab-lang-overlay is defined and not empty, also apply it.
• If tdd → Standard TDD rules apply
• If manual or j-settings.md missing → Default to tddab, load the /tddab-planner skill

Read the mindset file completely. The rules in that file are the ONLY source of truth for the review. Do NOT apply rules from memory or training — only what the mindset file says.

2. Find the Plan

• If a plan path was provided → use that
• Otherwise look in current directory for plan.md or index.md
• If not found → STOP with error: "Plan file not found. Provide the path."

3. Detect Plan Type

• If plan has <red> with - test: lines → TDDAB plan, use tddab-planner.md rules
• If plan has <actions> with - action: lines → Step plan, load the /step-planner skill for rules
• If plan has <files> tag → multi-file plan, read index.md + all sub-files

4. Review — Apply Rules From Mindset File

Do NOT use a hardcoded checklist. Review the plan against the rules you read in step 1. The mindset file defines what is correct and what is not.

Focus on these categories:

A. Structural Correctness

• Are all required tags present as defined in the mindset file?
• Are block IDs unique and correctly formatted?
• Is <mission> comprehensive enough for clean-context execution?

B. Dependency & Ordering (CRITICAL)

• For each block: does it use types, functions, or files defined in a LATER block?
• If yes → dependency error — the block cannot execute without the later block
• Verify the execution order matches actual dependencies
• Check "no dependencies on future blocks" rule

C. Self-Sufficiency

• Can each block be understood with ZERO context beyond the mission?
• Are file paths complete?
• Are there references to "previous discussion" or "as we decided"?

D. Completeness (as defined by mindset file)

• Check what the mindset file says about code completeness — apply THOSE rules, not stricter ones
• Check for TODOs, unresolved decisions, "Option A or B"
• Check for "..." that skips DECISIONS (not boilerplate — boilerplate "..." is fine per mindset)

E. Project Conformity

• Follows project architecture patterns
• No security issues (SQL injection, XSS, etc.)

F. Requirement Coverage (BLOCKING — only if requirements.md exists)

This is the gate against all-or-nothing near-misses, where the plan satisfies most of the task but silently drops one requirement. If requirements.md exists:

• Exhaustive list: re-read instruction.md clause by clause and confirm requirements.md captured EVERY clause — especially secondary ones buried in compound sentences (negative "must not", disabled/read-only states, restore-on-switch-back, "exactly once", "no duplicates", exact response shapes). A missed clause is a BLOCKING gap — add the requirement, then fail.
• Every requirement is tested: every R1..Rn must map to at least one <red> test in the plan (via the test annotations / the plan's coverage lines). Any R with zero covering tests is a BLOCKING gap — add the block/test, then fail.
• Surface coverage: for any requirement that ranges over a category (every control / callable / endpoint / field / UI surface), confirm the plan tests EACH enumerated member, not a representative subset — especially members with different mechanics (native control vs custom dropdown/picker vs file input). A category requirement covering only some members is a BLOCKING gap.
• Do not pass the plan while any requirement is uncovered, any instruction clause is unmapped, or any category requirement covers only a subset of its members.

5. CVM Structural Validation (if available)

If mcp__cvm__parsePlan tool is available, call it on the plan file as a final objective check. The parser validates tag structure, block IDs, and format — things Claude's review might miss.

• If parsePlan returns errors → add them to the report as Structural Issues
• If parsePlan succeeds → note "CVM parsePlan: valid" in the report
• If CVM is not available → skip this step (review is still valid without it)

6. Submit the verdict — ONE word only

The review itself (checking rules, listing issues, fixing plan.md) is your WORK, done on screen in your turn. What you SUBMIT is only the verdict — a single bare token:

• No issues found AND (if available) parsePlan valid → submit passed
• Any issue, parsePlan error, or doubt → fix plan.md first, then submit failed

Submit exactly passed or failed — lowercase, one word, nothing else. No report, no ✅, no summary. The fixes go into plan.md, never into the reply. Submit failed, not failed. dependency issue in block 03.

7. Update Plan (if needed)

If issues were found:

• Edit the plan file with corrections
• Re-run review from step 4

Rules

• The mindset file is the ONLY source of truth — do not invent stricter or looser rules
• Be strict on dependency ordering — this causes real compilation failures
• Be strict on self-sufficiency — blocks execute on clean context
• Be lenient on code style — the mindset file defines what "complete" means
• Explain WHY something is wrong
• Suggest specific fixes, not vague advice
• If plan is good, say so quickly and move on
• Gate output: submit ONLY the verdict word — passed or failed, lowercase, nothing else. No report in the submit. The review notes stay on screen in your turn. Submit passed, not passed. <reasons>.