research(aif-handoff-bridge): Sub-wave B2 — Variant B' (hybrid CC-hook + aif-handoff MCP) REFERENCE

[artyhoo]

Summary

Sub-wave B2 R-phase patch evaluates Variant B' (hybrid) of the aif-handoff-as-runtime-bridge umbrella per kickoff §6 SW-B2 (added 2026-05-29 post-Stage-1, after maintainer surfaced gap in original Variant B framing). B' is the hybrid where we build a small fs-watcher on OUR side (CC PostToolUse hook OR standalone Node/bash watcher) that calls aif-handoff's existing handoff_create_task MCP — explicitly distinct from SW-B's REJECT of native upstream fs-watcher.

Verdict: REFERENCE — design-vocabulary for dispatch-trigger optimisation of Variant A; not actionable standalone.

• CC PostToolUse hooks expose a purpose-built mcp_tool hook type for already-connected MCP servers (code.claude.com/docs/en/hooks.md fetched 2026-05-29) — T-AIF-BRIDGE-B'-1 evades.
• aif-handoff MCP server runs standalone without coordinator daemon for task creation (DeepWiki ×2 phrasings) — T-AIF-BRIDGE-B'-3 evades.
• 4 sub-options fit <100 LOC: Option (a') = 0 LOC bash (mcp_tool config), (a) = ~25 LOC bash + curl, (b) = ~15 LOC chokidar Node, (c) = ~10 LOC fswatch bash.

BUT: B' inherits all 3 of SW-A's ADOPT-blockers verbatim (PLAN.md disk coupling, WebSocket-broadcast no-topic-filter, autoMode Reviewer ↔ reviewer-discipline.md §2 conflict — DN-1=B-constrained applies). B' optimises only the dispatch trigger; the load-bearing blockers persist.

Falsifier: B' upgrades to ADOPT VOCABULARY iff Sub-wave D greenlights Variant A AND DN-1=B-constrained adopted. Comparison table vs A and B in §6 (Sub-wave D input).

Test plan

• R-phase only — no code, no skill edits, no SSOT row landings (output = single markdown research-patch per kickoff §11)
• Own cold-QA before handoff (T19) — file:line citations re-verified; SKILL.md substrate-purity ref corrected :338→:441 (drift since SW-A); inject-matching-rule.sh §2.1 content-read claim corrected to accurate parse-only-not-content shape
• CI green at push (153 passed / 4 skipped / 21 test files / 22.86s); markdownlint <600 LOC enforced
• §1.7 PR-body discipline applied below (CI gate WILL NOT fire on research-patches/** per dispatch instruction — pre-flight grep is the only programmatic check; H3 headers + word «applied» verified)

§1.7 Forward-check applied

• build-first-reuse-default.md §1 (verdict ladder): B' = REFERENCE — fs-watcher mechanism reuses existing CC hook precedents (SSOT #20 ADOPT for CC hooks API; .claude/hooks/inject-matching-rule.sh:80 precedent for path-scoped Edit|Write); aif-handoff MCP standalone-callability verified via dual-DeepWiki triangulation. BFR §3 6-layer search applied to this patch itself (§7 of the patch; §10 self-application).
• no-paid-llm-in-ci.md §1: All evidence via DeepWiki MCP (subscription-bundled), WebSearch (harness-included), WebFetch (harness-included), gh CLI (free), local Bash grep. Zero API-billed calls.
• reviewer-discipline.md §2: This patch does NOT pick project strategy. Where verdict touches Variant A or B' together (the «greenlight both» question), patch surfaces it as a Sub-wave D input — see §6 synthesis observation at docs/meta-factory/research-patches/2026-05-29-aif-handoff-bridge-variant-b-prime-hybrid.md#§6 (line 433).
• ai-laziness-traps.md §3: T1 / T3 / T7 / T11 / T12 / T13 / T15 / T16 / T17 / T19 / T20 + domain T-AIF-BRIDGE-B'-1..5 all enumerated and walked at ...variant-b-prime-hybrid.md §7 line 464.
• doc-authority-hierarchy.md §3: Header carries Status + Class (N/A — research-patch) + Authoritative-for + NOT authoritative-for (patch lines 5-6).
• dual-implementation-discipline.md §2: Patch is markdown-only artefact under §2 (i) markdown-only carve-out — rule does not apply at R-phase layer.
• DECISION=C invariant: §4 criterion 2 verified HARD-requirement satisfied across all 4 sub-options (patch lines 309-334).
• CLAUDE.md PR strategy: Strictly within Sub-wave B2 scope. No drive-by edits to SKILL.md, SSOT, install.sh, or any non-research-patches file. One file, one commit.

§1.7 Backward-check applied

• SSOT Wave 0.5: revise §13.21 L3 closure plan + self-review #20 (CC hooks API, ADOPT): Re-verified by direct WebFetch (§2.1) — not silently superseded. The mcp_tool hook type quoted here is recorded as a usage exemplar of Wave 0.5: revise §13.21 L3 closure plan + self-review #20, not as a new capability.
• SSOT docs(research-patches): Wave 6 AI-doc cold audit (§13.26) #27/docs(wave-6): close §13.26 — AI-doc effectiveness cold audit + review verdicts #28/feat: Wave 7 — hot-checks + harness-hooks + template test + §13.23 4th-layer #29/docs(wave-5): readiness REVISE fixes — SSOT ID remap + §13.25 status #30/Wave 8.4: pre-push §7 Prior-art escape-hatch substance arm #43/Arm Phase 10 (§13.32) + ai-laziness-traps project rule #44/docs(refactor): split open-questions.md into open + closed-questions archive #46/docs: post-1A follow-up (Class C reviewer-discipline, Q3/Q5, README absolutism) #67/feat(agents): resolve C-1 agent-name collision with AI Factory (KEEP-AIF / skill-context / rename) #80: None silently superseded. §9 of the patch proposes additive notes only (SSOT Arm Phase 10 (§13.32) + ai-laziness-traps project rule #44 records «B' hybrid evaluated separately; same DEFER unless A is greenlit»). No verdict changes.
• SW-A patch (2026-05-29-aif-handoff-bridge-variant-a-mcp.md): This patch cites SW-A's findings verbatim (the 3 ADOPT-blockers in §3 last 3 rows + §6 last 3 rows) and does not re-evaluate them. B' verdict explicitly depends on SW-A verdict per §5 falsifier.
• SW-B patch: Evaluates a different mechanism (our-side fs-watcher vs upstream native fs-watcher). SW-B's REJECT stands; B' is a separate evaluation per kickoff §6 SW-B2 origin.
• No .claude/rules/* / .claude/skills/* / agents/* / packages/* / install.sh / SKILL.md / settings.json modified. Sketch JSON/bash/Node fragments in §4 are illustrative only — NOT shipped.
• Scope: Variant B' only. Variants A/B/C/D have separate Sub-waves. No scope creep.
• T15 self-application: Patch applied BFR-default §3 to itself — SSOT consult (rows Wave 0.5: revise §13.21 L3 closure plan + self-review #20 + docs(research-patches): Wave 6 AI-doc cold audit (§13.26) #27/docs(wave-6): close §13.26 — AI-doc effectiveness cold audit + review verdicts #28/docs(wave-5): readiness REVISE fixes — SSOT ID remap + §13.25 status #30/Wave 8.4: pre-push §7 Prior-art escape-hatch substance arm #43/Arm Phase 10 (§13.32) + ai-laziness-traps project rule #44/docs(refactor): split open-questions.md into open + closed-questions archive #46/docs: post-1A follow-up (Class C reviewer-discipline, Q3/Q5, README absolutism) #67/feat(agents): resolve C-1 agent-name collision with AI Factory (KEEP-AIF / skill-context / rename) #80), DeepWiki ≥2 dual-phrasing triangulation, WebSearch ≥3 phrasings, WebFetch primary-source on CC docs, own-stack precedent inspection. See patch §10 line 553.

---

Prior-art: SSOT #20 (CC hooks API, ADOPT) + SSOT #27/#28/#30/#43/#44/#46/#67/#80 (aif-handoff rows reviewed). New row proposed at slot #85 (next-available per grep on prior-art-evaluations.md).