RF — TFW-21: Research Workflow Compression
Date: 2026-04-03
Status: 🟢 RF
Changes
| # |
File |
Action |
Description |
| 1 |
.tfw/templates/RES.md |
Modified |
Added 3 checkpoint fields (Agent assessment, Depth check, Recommendation) to all 3 stage checkpoints. Added external research line to Sufficiency Check self-check |
| 2 |
.tfw/workflows/research.md |
Rewritten |
Compressed from 2397→1145 words, 319→160 lines |
| 3 |
.agent/workflows/tfw-research.md |
Synced |
Exact copy of .tfw/workflows/research.md |
| 4 |
README.md |
Updated |
Task Board: TFW-21 status → 🟢 RF |
What Was Removed
| Section |
Words removed |
Reason |
| Example Flow (lines 199-243) |
~280 |
Template RES.md + Hard Rules sufficient |
| "What good research looks like" (lines 259-264) |
~60 |
Duplicates Hard Rules #1-#3 |
| "What bad research looks like" (lines 266-272) |
~60 |
Duplicates Anti-patterns |
| "Operational" (lines 274-278) |
~40 |
Repeats Role Lock |
| Anti-patterns duplicate block (lines 303-318) |
~150 |
Merged into Rules § NEVER |
| Inline Checkpoint format (lines 131-144) |
~70 |
Moved to templates/RES.md |
| Inline Sufficiency Check (lines 150-170) |
~100 |
Already in templates/RES.md |
What Was Preserved
- ✅ Research Mindset — full tone, compressed from 9→5 lines
- ✅ All 8 Hard Rules (MUST section)
- ✅ 3 stage mindset reminders ("> Remember: ...")
- ✅ Briefing Protocol with WAIT
- ✅ Closure Protocol (5 steps)
- ✅ Entry Modes, Prerequisites, Limits, Status Transitions
Verification
Before: 2397 words, 319 lines
After: 1145 words, 160 lines (-52% words, -50% lines)
Hard Rules (MUST): 8 ✅
Stage reminders (Remember:): 3 ✅
Briefing mentions: 4 ✅
Closure mentions: 3 ✅
Stage definitions: 3 ✅
WAIT stops: 3 ✅
Adapter diff: identical ✅
Observations (out-of-scope, not modified)
| # |
File |
Type |
Description |
| 1 |
.tfw/workflows/plan.md |
duplication |
Naming Rules table (lines 64-75) duplicates conventions.md §4 — ~100 words recoverable |
| 2 |
.tfw/workflows/handoff.md |
duplication |
Anti-patterns section partially overlaps conventions.md §14 |
Fact Candidates
| # |
Category |
Candidate |
Source |
Confidence |
| 1 |
constraint |
AI workflow instructions degrade at >~1200 words per document — agents lose mid-document attention. Working range: 700-900 words |
TFW-21 analysis + external research |
Medium |
| 2 |
convention |
Checkpoint behavioral fields (Agent assessment, Depth check, Recommendation) must live in templates/RES.md, not inline in workflow docs |
TFW-21 RES R1 |
High |
RF — TFW-21: Research Workflow Compression | 2026-04-03