trace-first-starter

RF — TFW-14: Research Interaction Model

Дата: 2026-04-01 Автор: Executor (AI) Статус: 🟢 RF — Выполнено Parent HL: HL-TFW-14 TS: TS TFW-14

#	File	Action	Changes
1	`.tfw/workflows/research.md`	MODIFY	Briefing Protocol (§ before Process), Stage Handoff (checkpoint extension), Closure Protocol (§ after Final Checkpoint), Sufficiency Check (replaces Complexity Check), 3 new Hard Rules (#5-7), 4 new Anti-patterns, per-stage→per-turn in Limits table + Hard Rule #4
2	`.tfw/templates/RES.md`	MODIFY	Briefing section (plan, scope intent, guiding questions) after Research Context; Closure section (HL recommendations table, next step) before Conclusion; Sufficiency Check replaces Complexity Check; Verdict wording updated
3	`.tfw/workflows/plan.md`	MODIFY	Phase 3.5 rewritten: pros/cons format, default=recommend research, HL update gate after RESEARCH, separate session recommendation; 2 new Anti-patterns
4	`.claude/commands/tfw-research.md`	MODIFY	Step 4 synced: Briefing → Stages → Closure structure, turn-based rhythm, Sufficiency for HL finalization
5	`.agent/workflows/tfw-research.md`	MODIFY	Step 4 synced: identical to Claude Code adapter

Budget: 0 new files, 5 modifications. Within scope.

#	TS says	What I did	Justification
1	TS does not mention Limits table (L183)	Updated `Questions to user per stage` → `per turn`	ONB Recommendation #1: direct consequence of turn-based rhythm change. Without this, Limits table contradicts new Hard Rule #4
2	TS does not mention Verdict wording (L135)	Updated `Sufficient for TS / Need another pass` → `Sufficient for HL finalization / Need another pass`	ONB Recommendation #2: Verdict line is part of the same Final Checkpoint block being modified
3	TS does not mention Hard Rule #4 (L151)	Updated `per stage` → `per turn`	ONB Risk #1: old wording directly contradicts the new turn-based rhythm

All deviations are consistency fixes within the same file, approved in ONB.

No automated tests. Manual verification:

✅ All 5 files modified as specified
✅ Briefing Protocol: 10 lines in research.md, anchored in RES.md template
✅ Stage Handoff: 2 lines added to checkpoint step (step 5 + sub-bullet)
✅ Closure Protocol: separate section with 4 steps + 2 callout blocks
✅ Sufficiency Check: self-check list (4 items) replaces Complexity Check
✅ Hard Rules: 7 rules (3 new: #5 briefing, #6 closure, #7 sufficiency)
✅ Anti-patterns: 14 total (4 new: skip-briefing, rush-bias, no-closure, skip-bias)
✅ RES.md: Briefing section (17 lines), Closure section (8 lines)
✅ plan.md: Phase 3.5 restructured, 2 new anti-patterns
✅ Adapters: both identical, 6 lines replacing 4
✅ Per-stage→per-turn consistency: Hard Rule #4, Limits table, Anti-patterns

No live validation. Changes are structural (markdown workflow), not code. The true test is the next /tfw-research session — will the agent follow Briefing and Closure? This requires a real task.
Standalone mode. TS explicitly excludes standalone research. Closure Protocol references HL update, which doesn’t apply to standalone. Current wording says “Research agent writes recommendations; coordinator applies them” — reasonable for both modes, but standalone may feel awkward.

#	File	Line(s)	Type	Description
1	`.tfw/workflows/research.md`	L27	style	`This process gives birth to the details needed for TS` — still references TS as the primary output. With Closure Protocol, the primary output is now HL update recommendations. Consider rewording to mention HL
2	`.tfw/glossary.md`	L60	naming	RESEARCH glossary entry says “Optional — user can skip with confirmation.” This is accurate but doesn’t mention the new pros/cons format or default recommendation. Minor sync opportunity
3	`.tfw/glossary.md`	L66	naming	Pass definition says “Minimum 1 pass required” — still references old model. With Sufficiency Check, the framing is now “sufficient for HL finalization” which is slightly different intent
4	`.tfw/conventions.md`	L50	naming	RES artifact description: “Living document: decisions and questions at the top, stage logs below” — doesn’t mention Briefing or Closure sections. Template now has them

*RF — TFW-14: Research Interaction Model

2026-04-01*

This site is open source. Improve this page.