metis-premier-proposal

Produce premier client-ready proposals, capabilities decks, and case studies for Metis Strategy BD work. This skill encodes lessons from eight iterations of the Citi CRS Proposal (where HTML + Playwright and slide-screenshot pasting both failed) and the AI PM Capabilities deck (which exposed the need for narrative-first planning and flexible layout variety). ReportLab with coordinate-based placement and a measure-first layout pattern is what actually works for fixed-page designed documents.

Output Specification

• Format: PDF, 16:9 landscape
• Dimensions: 13.33" x 7.5" (matches PowerPoint slide aspect ratio)
• Typical length: 10–25 pages (mode-dependent)
• File size target: 5–15 MB
• Engine: ReportLab with coordinate-based placement
• Content source: YAML file (flat sections: list) + Python dispatch

Pick Your Mode

Three modes, three starter scripts. Mode shapes the entire document: close discipline, language, what to include, what to omit.

Mode differences matter:

• Language mirroring is a hard rule only in buyer mode. Use the buyer's transcript vocabulary verbatim in titles and pull quotes. Capabilities mode uses industry-neutral language. Case-study mode mirrors the engagement's language.
• Problem framing differs: buyer mode anchors in the buyer's own quoted pain; capabilities anchors in market dynamics with industry-typical metrics; case-study uses setup/situation, not problem framing.
• Closing differs: buyer ends on CTA, capabilities on contact block, case-study on credit plus what-good-looks-like.

Workflow

1. Gather Inputs

Before any planning, collect:

• Transcript of any discovery call (buyer mode). Extract the buyer's actual words and phrasings.
• Source material such as prior proposal decks, relevant client case studies, and Metis IP decks.
• Industry context — financial services, healthcare, technology. This drives case study selection.
• Named stakeholders mentioned in the conversation (partners, blockers, champions). Verify spelling.

2. Consultant Planning Phase — Brief Before Draft

Do not start writing YAML until you have a thesis and a sketch. Jumping to YAML first is the single biggest quality trap.

Fill in scripts/_brief_template.md in the project's working folder, renamed to brief.md. The brief captures: audience, thesis, three-sentence summary, signature visual, three golden proof points, natural cleavage, narrative sketch, pacing map, kill list.

Every checkbox in the brief must be ticked before YAML drafting begins. See references/narrative-planning.md for the full planning loop (audience → thesis → signature visual → source triage → natural cleavage → sketch → reverse-outline → pacing → kill list).

Source triage and kill-your-darlings disposition: Read all source material first. Extract the 3–5 details that carry the thesis. Cut everything else, even the good stuff. A tight deck that lands beats a complete deck that doesn't. The skill's scaffolding implicitly encourages filling slots; subtract on purpose.

3. Write YAML Content File

Every piece of editable text goes in a flat sections: list. Each section declares:

- type: pillars                    # dispatch key → renderer
  intent: "Show we own PM, not advise on it"  # one-line "so what?"
  eyebrow: "Module B"
  title: "Four Pillars of Embedded Product Management"
  # ...content_type-specific fields

• type is required. Missing or unknown types raise an error with the valid list.
• intent is required on all content pages (covers, dividers, and closings are exempt). It does not render. Its purpose: force the author to name the job of the page before drafting copy. At polish-pass time, re-read each page and ask: "does the copy actually do what intent says?"

Valid content types: cover, problem_framing, section_divider, modules_overview, activities_deliverables, pillars, timeline, signature_image, quote, comparative, metric_dense, minicase_pair, proof_overview, next_steps, closing. See references/page-patterns.md for each.

Module count follows the story, not the template. The modules_overview helper auto-picks the grid:

• 3 modules → wide 3-column row
• 4 → 2×2 (current Citi default)
• 5 → 3-top + 2-bottom staggered
• 6 → 3×2
• 7+ → raises ValueError

Each module typically gets 2–3 pages. Do not pad a 1-page idea to 3 pages because other modules have 3. See references/content-rules.md structural consistency section.

4. Extract and Crop Slide Images (if needed)

If embedding visuals from an existing PPTX:

1. Export slides as PNG using scripts/crop_slide.py or PowerPoint's native export
2. Crop out the slide's own title, subtitle, and footer — otherwise the rendered page will have duplicate titles
3. Save cropped versions with a -cropped suffix in the project's _working folder

Use the signature_image content type to render cropped visuals as a hero page with caption and callout.

5. Build with ReportLab (Measure-First Pattern)

The core technical rule: measure content before placing containers. See references/architecture.md. Helpers in scripts/helpers.py implement this pattern for bullet lists, callouts, cards, metric strips, and all the new pattern renderers.

Run the appropriate starter:

PYTHON="C:/Users/.../python.exe"
"$PYTHON" starter_buyer.py my-proposal-copy.yaml
# or starter_capabilities.py / starter_casestudy.py

The starter dispatches each section to its renderer and warns on missing intent: or three-in-a-row layout repetition.

6. Mechanical QA

Two verification passes:

# Overlaps + content rules + metadata
"$PYTHON" scripts/verify.py --pdf output.pdf --mode all

# Layout repetition (no pattern 3+ in a row)
"$PYTHON" scripts/verify.py --pdf output.pdf --mode repetition

# Reverse-outline: titles only, in sequence
"$PYTHON" scripts/verify.py --pdf output.pdf --mode titles

# Visual — render every page and eyeball it
"$PYTHON" scripts/verify.py --pdf output.pdf --mode render --out verify/

Never declare the PDF done without visual verification. Bbox analysis alone missed several critical issues in the Citi engagement.

7. Polish Pass — Three Passes Before Shipping

After mechanical QA passes, run the polish pass. See references/polish-pass.md.

• Pass 1 — Does it land? Read the full deck once, cold. Do you believe the thesis by page 5?
• Pass 2 — Is every word pulling weight? Read each page slowly. Cut 10%. Adjective audit. Re-read intent and ask: "does the copy do that?"
• Pass 3 — Reverse-outline. Run --mode titles. Read titles in sequence. Do they tell the story without the body copy? If yes, ship. If not, fix structure, not copy.

8. QA Checklist (Content Rules)

Before declaring content done (full list in references/content-rules.md):

• No em dashes (—, \u2014)
• No contractions in formal body text
• No absolute claims ("Zero", "Always", "Never", "Every")
• No named products from other client engagements
• Stakeholder names spelled correctly
• Transcript quotes appear at most twice
• Buyer's actual phrases in titles/subtitles/callouts (buyer mode)
• Case studies match prospect's industry or are industry-neutral
• Module page counts honest (2–3 typical, no padding)
• No layout pattern repeats 3+ times in a row
• Every module page has an intent: field
• Each callout label is 1–4 words, all caps
• Metric cards use round numbers with "+" or ranges
• Next Steps page has exactly 3 steps (buyer mode only)

9. Deploy

• Keep all working files (unpacked source, cropped images, intermediate PDFs, build scripts, brief.md) in G:\My Drive\Vault\_working\<project>\
• Copy only the final PDF to the shared drive: G:\Shared drives\BizDev Collaboration\<Client>\<Year>\ (buyer mode) or the appropriate marketing folder (capabilities/casestudy)
• The vault's global CLAUDE.md has the rule: shared drives are for deliverables only, never intermediate artifacts.

References

• references/narrative-planning.md — Consultant planning loop (audience → thesis → sketch → kill list)
• references/polish-pass.md — Three-pass critical re-read after mechanical QA
• references/architecture.md — Why ReportLab + measure-first pattern + YAML separation
• references/page-patterns.md — All 15 layout templates with code snippets
• references/brand-standards.md — Metis brand colors, fonts, logos, graphic devices
• references/content-rules.md — Editorial standards (no em dashes, no absolutes, intent per page, layout repetition, balance)
• references/qa-process.md — Verification workflow
• references/failure-modes.md — Things that broke during builds and how they were fixed

Scripts

• scripts/helpers.py — Reusable layout helpers (measure_body_text, metric_card, activities_deliverables_page, section_divider with 3 styles, modules_overview_page auto-grid, and six pattern helpers: minicase_pair_page, signature_visual_page, process_timeline_page, quote_led_page, comparative_page, metric_dense_page)
• scripts/verify.py — Overlap, content, repetition, titles, render, metadata, and all-in-one checks
• scripts/crop_slide.py — Crop slide screenshots to remove titles/footers
• scripts/starter_buyer.py — Buyer-mode dispatch starter (+ starter_buyer_copy.yaml template)
• scripts/starter_capabilities.py — Capabilities-mode dispatch starter (+ starter_capabilities_copy.yaml template)
• scripts/starter_casestudy.py — Case-study-mode dispatch starter (+ starter_casestudy_copy.yaml template)
• scripts/_brief_template.md — Pre-build brief template (fill in before YAML)
• scripts/starter_build.py — DEPRECATED. Kept as a shim for legacy nested-key YAMLs.

Reference Implementations

• Buyer mode: Citi CRS Proposal (April 2026), Andrew Krusell to Amresh Mathur.
  • G:\My Drive\Vault\_working\citi-crs-proposal\citi-proposal-copy.yaml (legacy nested format)
  • G:\Shared drives\BizDev Collaboration\Citi\2026\Citi CRS Proposal - Metis Strategy 2026-04-15.pdf
• Capabilities mode: AI PM Capabilities (April 2026).
  • G:\My Drive\Vault\_working\ai-pm-capabilities-proposal\ai-pm-capabilities-v2.yaml (new flat sections format — canonical example of the post-refactor shape)
  • AI-PM-Capabilities-v2_Metis-Strategy.pdf (22-page deck)

Activation

When this skill is triggered, the first response to the user should confirm:

1. Mode selection (buyer / capabilities / casestudy)
2. Understanding of the client and context
3. The planned narrative sketch — not the page structure, but what the document argues
4. The content sources you will use
5. Whether you need any additional inputs (transcripts, case studies, stakeholder names)

<<<<<<< HEAD Never start coding the PDF before this plan is confirmed. The Citi engagement lost multiple days to rebuilds because the approach was not settled upfront.

======= Then draft brief.md in the working folder. Never start coding the PDF before the brief is approved. The Citi engagement lost multiple days to rebuilds because the approach was not settled upfront. The AI PM deck exposed the same failure mode: an unintended four-module shape propagated from the template because the narrative was not planned first.

fdd4c4e (Refactor metis-premier-proposal: narrative-first planning, flexible layouts, three modes)