v2.6.1 — Word doc design fix

What changed

Design audit of actual Word doc output vs reference doc. Fixed 7 root causes of cramped, amateurish layout.

Key fix: Field labels inside gap cards had only 80 twips of space before their content — the "squished headlines" problem. Now 140 after label + 200 before subsequent labels = clear visual hierarchy throughout.

All changes are in the Word doc build protocol only. Report output, all modes, routing, memory — untouched.

Full fix list

• Field label after 80 → 140; subsequent labels before 0 → 200
• Gap name header cell: added explicit line spacing
• Gap card body bottom margin 280 → 360
• Question sub-blocks before/after 80 → 120
• Spacer between gap cards 280 → 480
• Section title after 200 → 320
• Script phase label before 200 → 280, after 80 → 120
• Script intent note: explicit before:0, after 100 → 140
• Script quote block after 200 → 280
• Admissions context note before 0 → 80, after 120 → 160
• What to Avoid before/after 80 → 120
• MEDDPICC header top/bottom 100 → 140, left/right 150 → 200
• MEDDPICC data rows top/bottom 80 → 100, left/right 150 → 200
• Blind spots before 160 → 240, after 80 → 120

v2.6.0 — Self-Contained Word Doc Build

What's New

Word doc build: completely rebuilt. Self-contained. No external dependencies.

Bug 1 — External skill dependency (fixed)

The previous protocol told the agent to invoke anthropic-skills:docx via the Skill tool. This was unreliable — it stalled, failed silently, or required 2–3 follow-up prompts before it executed. The agent now writes a complete Node.js script using the docx npm package and runs it directly with node. One shot. Nothing external. Node is always present in Claude Code environments.

Bug 2 — Invisible cover page (fixed)

XML inspection of generated documents confirmed the cover page had white text with no background color — white text on white paper, invisible. The cover is now built as a full-width table cell with navy shading (#1a2744), which guarantees the background renders correctly.

Design improvements

• Gap card side padding doubled: 200 → 400 DXA (~10pt → ~20pt)
• Gap card vertical cell padding increased to 240/280 DXA
• Section title spacing increased for breathing room
• Script quote blocks wider indent on both sides
• Explicit spacer paragraphs between cards (instead of relying on table margins)
• Design mandate added to protocol: document must feel spacious — if cramped, increase padding

What didn't change

Report content, analysis quality, all 9 sections, every other mode (discovery, review, debrief, email, scope feedback), memory system, routing logic — untouched. The change is scoped entirely to the Word doc build path.

---

Gapsi is built on peer-reviewed behavioral science: Kahneman & Tversky (Prospect Theory, 1979), Tversky & Kahneman (Framing Effects, 1981), Samuelson & Zeckhauser (Status Quo Bias), Kahneman, Knetsch & Thaler (Endowment Effect, 1990), Heath, Larrick & Wu (Goals as Reference Points, 1999), Novemsky & Kahneman (Boundaries of Loss Aversion, 2005).

v2.5.0 — Word doc build protocol: one-shot execution + designed layout

What changed

Added ▸ WORD DOC BUILD PROTOCOL — a dedicated section that tells the agent exactly how to build the Word doc in one shot.

Fixes two bugs:

Bug 1 — Stalling on "yes"
The agent was invoking anthropic-skills:docx as a background delegated task and then monitoring for completion — requiring 2–3 prompts before it executed. The protocol now mandates: invoke the skill immediately via the Skill tool, no acknowledgment message, no monitoring, no check-ins. One shot.

Bug 2 — Plain, undesigned output
The doc was flat white text with no visual hierarchy. The protocol now specifies the complete layout:

• Cover page — dark navy background, white text, deal name + call type + date + rep + "Prepared by Gapsi"
• Page breaks between all 8 major sections
• Gap blocks — bordered navy cards with light blue fill, labeled fields inside
• Script lines — shaded grey blocks per exchange
• Admissions — left-bordered callouts with cream background
• The One Sentence to Remember — centered pull quote between two navy rules
• MEDDPICC — three-column table with color-coded status symbols (✓ green / ~ amber / ? red)
• What to Avoid — red-tinted callouts with dark red left border

What did NOT change

Report output format (all 9 sections), every other mode, routing logic, memory system — untouched.

v2.4.0 — auto-output full report on transcript paste, no menu

What changed

The A–G output menu is gone.

Previously, pasting a transcript triggered a menu asking users to pick a letter (A through G) before seeing anything. Most users don't know what "MEDDPICC Deal Status," "Exact Language Pack," or "Gap Inventory" mean — and they shouldn't have to.

Now: paste a transcript → get the complete analysis immediately. Full report, no gate.

The report already contains everything the menu was hiding:

• Named business gaps with exact questions missed and solution fit
• MEDDPICC gaps surfaced inside the main analysis (not a separate option)
• Gaps to build on the next call with five-field structure
• The next call script — exact language throughout
• What to avoid + the one sentence to remember
• Decision maker status at the bottom

One question at the end: Want this as a formatted Word document? Say yes and I'll build it.

That's it. That's the only choice.

v2.3.0 — trim dead weight + MEDDPICC gap flagging

What changed

Cut A — Stale coaching output blocks removed

Removed 5 coaching output format blocks from Specialist Agents 1–5 (the Reference Point / Current Reality / Gap Calculation / Inaction Cost / Positioning format blocks). These were the v1.x report format. The deal-centric CALL ANALYSIS REPORT FORMAT introduced in v2.2.0 drives all transcript output — nothing routes to the old blocks.

Cut B — Script Rebuilder removed

Removed the Script Rebuilder section (~125 lines) and all references to it: the routing table row, Mode Menu option 9, and the description field trigger. The Script Rebuilder violated the analyst posture rule — it required two questions before doing any work, a confirmation pass before rebuilding, and per-gap confirmations throughout. Incompatible with output-first design.

MEDDPICC gap flagging

The Transcript Analysis Engine now checks Economic Buyer and Champion status after running the 5-vector analysis (new Step 3a). If either is unknown or partial, named gaps appear in the main analysis — not just the status block at the bottom of the report:

• The Decision Maker Visibility Gap — surfaces when Economic Buyer is unknown or the rep never asked who holds final signing authority
• The Internal Champion Gap — surfaces when no internal advocate has been identified

Both appear in GAPS FAILED TO CREATE and GAPS TO BUILD ON THE NEXT CALL with full structure: what happened / questions missed / the gap / solution fit / inaction cost. These are treated as business-critical gaps, not MEDDPICC bookkeeping.

v2.2.0 — Deal-centric report redesign

What Changed

The Call Analysis Report is now a full deal playbook, not a framework audit.

Core change: gaps are named after the business problem

Instead of "Reference Point — Weak," you get "The Aging Investor Base Gap" or "The Referral Dependence Gap" — named after what the gap actually is in this deal.

New report structure (A → Full Analysis):

1. The Core Principle — what the rep is actually trying to do, reframed as loss prevention for this prospect
2. What Rep Missed — verdict on the call + flags
3. Gaps Failed To Create — 5–8 named business gaps, each with: what happened / exact questions missed / the gap in one sentence / how offer fits
4. Gaps To Build On Next Call — same gaps as prep material: current state / desired state / gap / inaction cost / solution fit
5. Admissions Captured — every admission with context
6. The Next Call Script — exact language throughout (opening reframe, goal, reality, gap statement, inaction, offer reframe, price reframe, champion close, decision question)
7. What To Avoid — specific to this prospect
8. The One Sentence To Remember — the close question, deal-specific
9. MEDDPICC Status — blind spots and priorities

Output menu (new)

After pasting a transcript, Gapsi presents a menu before running any output:

• A — Full Analysis
• B — Gap Inventory only
• C — Next Call Script only
• D — Exact Language Pack
• E — MEDDPICC Status
• F — Offer vs. Gaps
• G — Full Analysis as Word doc

v2.1.0 — McKinsey-Style Report

What changed

Call Analysis Report — full restructure.

Old: flat list of 5 vectors in prose.

New: structured document report —

1. Executive Summary — verdict in 2-3 sentences before any detail
2. What You're Not Seeing — gaps ranked CRITICAL / SIGNIFICANT / MINOR with deal-specific findings
3. Full Analysis — the 5-vector detail, each in its own sub-section
4. Offer vs. The Gaps — how to reframe and reposition the offer specifically for this prospect based on what was found
5. Priority Fix — one sentence, maximum leverage
6. Recommended Next Step
7. Docx offer — every report ends with "Want this as a formatted Word document? Say 'docx' and I'll build it."

Reports are written in analyst voice — specific to the transcript, specific to the prospect, not generic coaching notes.

v2.0.0 — Analyst Posture Redesign

What changed

Onboarding — 6-step interrogation replaced with a single intake prompt. Suggests voice dictation. Gapsi reads the dump, organizes it, confirms in plain prose. Done in 2 exchanges instead of 6.

Transcript analysis — Zero questions before the report. Engine reads the transcript, extracts company/contact/call-type directly, creates the deal file automatically, outputs the full analysis immediately. No "which deal is this?" No "what type of call?" Just the report.

Format — All ASCII box-drawing characters removed throughout (╔══╗, ┌──┐, etc.). Every output now uses markdown headers, bold labels, and --- dividers. Readable like a document, not a terminal printout.

Analyst posture — Explicit mandate added: default to output, not questions. Give the analysis before asking anything.

v1.5.0 — Comprehensive README

Complete rebuild of the README as a full product narrative — built to sell the methodology to anyone landing on the repo.

New sections:

• The problem it solves (why deals actually stall)
• The science — all 6 peer-reviewed studies with their numbers, plus a prospect-theory chart
• The 5-step Loss Aversion Gap Framework, diagrammed
• Full-system mind map of every mode, protocol, and doctrine
• Orchestrator routing flowchart + the 9 modes
• How a deal moves through Gapsi (discovery → review → close lifecycle map)
• 6 real-world scenarios + a worked one-deal example as a sequence diagram
• Results per call / per deal / per quarter
• Benefits → second-order benefits → compounding benefits, diagrammed
• Daily and weekly workflow integration with time-cost table
• Coaching philosophy, privacy model, who it's for (and who it's not)

9 Mermaid diagrams, all parse-validated against Mermaid v11 — they render natively on GitHub.

No functional changes to the skill, hook, or installer.

v1.4.1 — README banner

Adds the Gapsi banner image to the README. No functional changes.