qa-review-bl

name	qa-review-bl
description	[QA Method] Triangulate each BL invariant against docs + live + source code, auto-apply confirmed changes to business-logic.md, and reconcile test-case coverage. Delegates the live axis to qa-testing-expert; runs the triangulation via ba-system-analyzer.
argument-hint	all | domain <name> | BL-<ID> | diff [--dry-run]
disable-model-invocation	true

/qa-review-bl — BL Invariant Triangulation Review & Auto-Apply

Audit the Business Logic oracle (.claude/knowledge/oracles/business-logic.md) against reality. For each in-scope BL-* invariant, gather evidence from three independent axes — docs (VirtoOZ MCP via /vc-docs), live (playwright via qa-testing-expert), source code (GitHub MCP on VirtoCommerce repos) — assign a verdict, and auto-apply confirmed changes to the oracle. Then reconcile test-case Business_Rule coverage so cases follow the corrected invariants.

This is the mechanism that keeps the oracle honest. The oracle is the shared contract the whole QA pipeline judges against (violations auto-flag bugs, tests map assertions to {BL}, fixes must "preserve BL-*"), so drift here is high-blast-radius. Auto-apply is gated by a strict evidence bar, not by a human (see Rules): a change lands only when all three axes produced concrete, agreeing evidence.

Usage

/qa-review-bl all                    # Audit every invariant (large — batch by domain)
/qa-review-bl domain cart            # Audit one domain (BL-CART-*)
/qa-review-bl BL-CART-010            # Audit a single invariant
/qa-review-bl diff                   # Audit invariants whose Source anchor changed, or promoted since last audit
/qa-review-bl domain cart --dry-run  # Triangulate + verdict, but write NOTHING (preview)

Supporting Files

• bl-audit-criteria.md — the evidence bar per axis, which VirtoOZ tool / which repo to use per domain, the verdict decision table, and the edit-safety rules.

Verdict taxonomy (per BL)

Verdict	Meaning	Action
CONFIRMED	3 axes agree; Rule text is accurate	No Rule change; auto-apply a missing/refreshed Source: or Verify: only
DRIFT	3 axes agree with each other but the Rule/Verify text is stale	Auto-apply corrected Rule/Verify + Amended: stamp
MISSING	Behavior is documented and coded and live, but no BL exists	Auto-apply a new BL-<DOMAIN>-<NNN> (next free ID; body only)
CONTRADICTORY	Axes disagree (e.g. docs say X, live shows Y)	NOT confirmed → reports/ba/bl-proposals-<date>.md for human
UNGROUNDED	≥1 axis produced no evidence	NOT confirmed → proposals file
STALE/RETIRE	Behavior removed everywhere	Draft a retire proposal → proposals file (retiring is destructive; stays human-gated)

"Confirmed" = CONFIRMED / DRIFT / MISSING with unanimous, evidenced triangulation. Everything else routes to the proposals file. That is not a human gate on confirmed items — it is the definition of "not confirmed."

Execution

Step 0: Load references + collect the deterministic inventory

1. Read bl-audit-criteria.md (this folder) and the oracle business-logic.md.
2. Run the deterministic parser/linter — it is the single source for structure + coverage:

   npm run bl:audit:collect                       # JSON: every invariant + fields + coverage + structural findings
   npm run bl:audit:collect -- --filter=BL-CART   # one domain
   

   scripts/knowledge/lint-bl.ts parses all 149 invariants into structured fields, flags structural issues (BLL-001 dup ID, BLL-002 bad severity tag, BLL-003 missing required field, BLL-004 misfiled prefix, BLL-005 sequence gaps), and cross-references regression/suites/** (BLC-002 a suite cites a BL absent from the oracle; BLC-004 an invariant no test case covers). Use its invariants[] as the work-list and its findings[] to seed the audit (a BLC-002 is a candidate rename/MISSING; a BLC-004 is a coverage gap for Step 4).

Step 1: Triangulate each in-scope invariant — PARALLEL fan-out (ba-system-analyzer)

Triangulation is read-only and per-invariant, so run it in parallel. Split the in-scope invariants into disjoint batches (by domain, then chunk) and dispatch up to 3 ba-system-analyzer agents concurrently (one Agent-tool call per batch, all in a single message — matches the 3-slot browser pool, .claude/rules/agents.md "batch in groups of 3"). Each parallel agent gets its own isolated browser slot (playwright-firefox / playwright-chrome / playwright-edge — never shared), and a distinct test/org user if the live axis needs auth (shared org cart contaminates — feedback_concurrent_runners_distinct_org_users_taskstop). A parallel agent gathers evidence + assigns a verdict + returns the proposed edit only — it does NOT write to business-logic.md (that is the serialized Step 3).

For every BL, each agent captures the three axes with concrete evidence, never a bare opinion (see bl-audit-criteria.md for the per-domain source map):

• Docs axis — /vc-docs (VirtoOZ MCP). Pick the topic-scoped tool by domain (e.g. StorefrontUserGuide/StorefrontDeveloperGuide for cart/checkout UX, PlatformDeveloperGuide for platform/admin, *SourceCode for "where is this implemented"). Capture a quote + doc reference.
• Source axis — GitHub MCP search_code / get_file_contents on org:VirtoCommerce (read-only; QA never clones). Capture a file:line anchor and the relevant code shape.
• Live axis — delegate to qa-testing-expert (playwright-firefox) to OBSERVE the behavior via the real UI/API. Capture an {OBSERVED} result + screenshot. Honors the REAL-USER rule — no browser_evaluate/run_code_unsafe bypass.

Step 2: Assign a verdict

Apply the taxonomy above using the decision table in bl-audit-criteria.md. A verdict of CONFIRMED/DRIFT/MISSING requires an evidence tuple from all three axes that agree; any missing or conflicting axis ⇒ UNGROUNDED / CONTRADICTORY.

Step 3: Apply policy — SINGLE-WRITER fan-in

Collect the verdicts from all parallel agents, then apply serially, one entry at a time, in this one orchestrator process. Concurrent writes to business-logic.md race and corrupt the file — the parallel agents returned proposed edits (Step 1), they did not write. For MISSING, re-read the current max BL-<DOMAIN>-NNN immediately before each insert so two parallel-discovered new invariants can't claim the same ID.

• CONFIRMED / DRIFT / MISSING (unanimous, evidenced) → auto-apply to business-logic.md:
  • Edit the entry body only — never rewrite the meta Severity-Tags table (a separate edit if ever needed).
  • Stamp - **Amended:** <date> (auto-applied, triangulated — BL-AUDIT-<date>) and refresh - **Source:** (file:line + docs ref).
  • For MISSING, assign the next free BL-<DOMAIN>-<NNN> (read the oracle for the max), place under the correct ## Domain heading, keep evidence env-agnostic (no env names/URLs/slugs — say "the environment").
• CONTRADICTORY / UNGROUNDED / STALE-RETIRE → write to reports/ba/bl-proposals-<date>.md (the existing draft format) for human decision. Do NOT edit the oracle.
• --dry-run → compute verdicts + the intended diff, write NOTHING.

Step 4: Reconcile test-case coverage

Feed the audit back into the test-cases review flow (this is the "review BL with the test-cases review flow" tie-in):

• BLC-002 (a suite cites a BL that doesn't resolve) and any ID renamed/added this run → run /qa-review-tests suite <ID> on the affected suites and apply the Business_Rule remap via /qa-review-tests --fix (which asks before writing CSVs).
• BLC-004 (an invariant no case covers) → note as a coverage gap for /qa-test-lifecycle Phase 3 (generation). Do not fabricate cases here.

Step 5: Write the audit report

Write reports/knowledge/BL-AUDIT-<date>.md (see .claude/rules/reports.md — knowledge-maintenance artifact, monitoring-summary size discipline, target 15–40 / cap ~100 lines):

• Per-BL verdict table (BL-ID, verdict, 3-axis evidence refs, action taken).
• Applied section: the entries auto-edited/added, with a one-line before→after each (full diff is in git diff).
• Not applied section: CONTRADICTORY/UNGROUNDED/RETIRE items → link to the proposals file.
• Coverage reconciliation summary (BLC-002 remaps, BLC-004 gaps).

Rules

• Auto-apply is gated by evidence, never by silence. A change lands ONLY as CONFIRMED/DRIFT/MISSING with concrete, agreeing evidence from all three axes. No evidence on an axis ⇒ not confirmed ⇒ proposals file. (This deliberately replaces the former "never auto-edit business-logic.md / human per-entry approval" rule; see the memory update in the design.)
• Parallel fan-out, single-writer fan-in. Triangulate in parallel (≤3 browser agents, disjoint batches, isolated sessions), but apply to business-logic.md from one serialized writer (the orchestrator). Parallel agents return proposed edits; they never write the oracle themselves — concurrent writes race and corrupt it.
• Body-only edits. Never rewrite the Severity-Tags meta table as a side effect (feedback_bl_promotion_table_separately).
• Env-agnostic (feedback_bl_oracle_env_agnostic). No env names, URLs, or slugs in any applied entry — even in a Source:/evidence note.
• Reversible. Every applied edit is recorded in the BL-AUDIT report and lives in a git-tracked file; keep edits minimal and per-entry so a single one can be reverted.
• Retiring is destructive → always a human proposal, never auto-applied.
• P0-security invariants clear the same evidence bar as any other — but given blast radius, if the live axis cannot safely be observed (e.g. a real privilege-escalation probe), treat the axis as absent ⇒ UNGROUNDED ⇒ proposals file.

Agent Delegation

Situation	Agent	Browser
Triangulation batch (docs + source + live + verdict) — up to 3 in parallel	ba-system-analyzer ×N	one distinct slot each: playwright-firefox / playwright-chrome / playwright-edge
A single complex/high-risk live repro the batch agent can't safely observe	qa-testing-expert	its own slot (sequential, not a 4th concurrent browser)
Apply to business-logic.md (Step 3)	the orchestrator (this skill) — single writer, serialized	—
Test-case Business_Rule remap (Step 4)	test-management-specialist via /qa-review-tests --fix	—

Concurrency cap: 3 browser agents total (.claude/rules/agents.md). In parallel mode each batch agent does its own live observation on its assigned slot — it does NOT additionally sub-delegate the live axis to qa-testing-expert (that would exceed the cap). Reserve qa-testing-expert for a follow-up single-invariant deep-dive, run sequentially. Each parallel agent uses a distinct browser session + distinct test user; never share.

Integration with Other Skills

• /qa-review-tests — the downstream reconciliation (Step 4). BL-002/BL-004 there are the judgment twins of this skill's BLC-002/BLC-004.
• /qa-test-lifecycle — runs this skill as its always-on BL-audit phase (4c), scoped to the BL-* a run surfaced (no opt-in flag — the old draft-only --update-bl step is retired); its Phase 6 G6 gate reads the audit outcome.
• /ba-analyze — the other producer of BL candidates; unconfirmed items from both flows share reports/ba/bl-proposals-<date>.md.