The Complete Production SOP: How to Deploy and Run Pitch Deck & Presentation Architect in Autonomous AI Agents
Step-by-step SOP to install and run Pitch Deck & Presentation Architect in Codex, Claude Code, and Cursor. Programmatic PPTX rules and self-healing pipelines.
Technical Overview & Architectural Context
Autonomous coding agents frequently stumble when tasked with generating slides. Left unchecked, language models default to wall-of-text bullet lists, default 4:3 canvas ratios, illegible font hierarchies, overlapping text frames, and missing contrast checks. When producing programmatic PowerPoint files via libraries like python-pptx, models routinely miscalculate coordinate transforms (Inches vs Pt), generate overflowing shape frames, and fail to calculate bounding boxes dynamically. The Pitch Deck & Presentation Architect skill resolves these systemic errors by imposing strict design boundaries: forcing a 16:9 widescreen canvas (13.333 x 7.5 inches), constraining cards to calculated grid systems, capping slides to one core message, and emphasizing high-impact metric callouts over bullet density. By packaging deterministic Python generation scripts and automated quality gates, the skill ensures that autonomous agents in Codex, Claude Code, Cursor, and Windsurf generate executive-ready presentations that pass brand and visual QA on the first invocation.
Prerequisites & Environment Requirements
Verified Search Intents & Long-Tail Queries Solved
This operational tutorial is engineered to provide deterministic solutions for these developer search queries:
Step-by-Step Execution Playbook
Local Environment Setup & SKILL.md Mounting
Mount the skill manifest into the runtime discovery directory to enable deterministic activation.
Autonomous runtimes like Claude Code, Cursor, and Codex parse .agents/skills or local project manifests to register operational boundaries and trigger matching. By placing the SKILL.md in your repository root or local agent directory, the agent loads procedural layout constraints prior to writing Python execution scripts.
mkdir -p .agents/skills/presentations && curl -fsSL https://askills.dev/api/skills/presentations/raw -o .agents/skills/presentations/SKILL.md---
name: presentations
description: Build high-impact PowerPoint (.pptx) and Google Slides decks with modern typography.
triggers:
- make pitch deck
- generate pptx
- slides for investor
- keynote presentation
dependencies:
python:
- python-pptx
- pillow
---
# Presentation Architect Rules
- Enforce 16:9 widescreen: prs.slide_width = Inches(13.333), prs.slide_height = Inches(7.5)
- One core takeaway per slide; max 3 highlight cards per grid layout.
- Never place raw text boxes without explicit Inches/Pt bounding boxes.
- Metrics must use 44pt+ bold text with muted 14pt category labels underneath.
- Maintain 4.5:1 text-to-background contrast ratio.Keep the SKILL.md file under 150 lines to prevent context bloat; let external Python helper templates handle complex SVG/shape geometry.
In Cursor and Windsurf, ensure the skill path is included in your project root or registered in workspace settings, otherwise local rule indexing will skip triggers.
Virtualenv Configuration & Dependency Validation
Verify isolated Python environment dependencies for python-pptx and Pillow to prevent execution breaks.
Agent tools invoke python scripts directly in the workspace shell. Pre-installing and validating python-pptx avoids runtime pip install failures within non-interactive sandboxes where network egress might be constrained or throttled.
python3 -m venv .venv && source .venv/bin/activate && pip install python-pptx pillowEnsure the active agent terminal session executes inside the targeted virtualenv by configuring Python interpreter path in .cursor/settings.json or setting CLAUDE_ENV.
python-pptx does not bundle LibreOffice for headless PPTX-to-PDF rendering; do not ask the agent to render PDF thumbnails unless LibreOffice CLI is explicitly present.
Intent Matching & Trigger Invocation
Invoke the agent using declared trigger phrases to ensure layout constraints are active.
Trigger matching works best when the prompt directly correlates with the registered triggers such as 'generate pptx' or 'make pitch deck'. This prompts the agent to follow the widescreen layout rules and dynamic metric card structures defined in the SOP.
Explicitly state slide dimensions and color palette tokens in the prompt to reduce iteration cycles.
Avoid vague prompts like 'summarize this report'; use 'generate pptx investor deck from this report' to guarantee trigger matching.
Verification, Linting & Output Quality Gate
Execute an automated self-healing validation script to verify PPTX layout metrics and shape boundaries.
Instruct the agent to inspect the generated PowerPoint file via python-pptx to confirm slide dimensions match 16:9 widescreen, all shapes lie within the printable canvas, and zero empty text frames exist.
python3 -c "from pptx import Presentation; prs = Presentation('output/deck.pptx'); assert round(prs.slide_width.inches, 3) == 13.333, 'Invalid aspect ratio'; print('Deck verification passed.')"Inspect git diff on the generating script rather than binary .pptx files to review structural changes made by the agent.
Do not commit generated binary PPTX files to git history; keep generation scripts versioned and deliverables ignored.
Real-World Scenario: Automating a Series A Pitch Deck with Key Metrics
“make pitch deck: Create a 5-slide Series A pitch deck for an autonomous developer tooling platform named 'AgentMesh' using python-pptx. Palette: Dark slate (#0F172A) background, Emerald (#10B981) highlights, Off-white (#F8FAFC) text. Slide 3 must feature 3 metric cards highlighting: $3.8M ARR (310% YoY), 140k Devs, and 99.4% Retention.”
Runtime Troubleshooting & Failure Modes
Root Cause: The agent passed a hex string (e.g., '#10B981') directly to the python-pptx font color setter instead of instantiating an RGBColor object.
from pptx.dml.color import RGBColor
def hex_to_rgb(hex_code):
h = hex_code.lstrip('#')
return RGBColor(*(int(h[i:i+2], 16) for i in (0, 2, 4)))
run.font.color.rgb = hex_to_rgb('10B981')Root Cause: The agent initialized Presentation() without overriding default 4:3 or standard dimensions before adding slides.
from pptx import Presentation
from pptx.util import Inches
prs = Presentation()
prs.slide_width = Inches(13.333)
prs.slide_height = Inches(7.5)Frequently Asked Questions (Pitch Deck & Presentation Architect SOP)
Q:Can I use Pitch Deck & Presentation Architect across Codex, Claude Code, and Cursor simultaneously?
Yes. The skill conforms to the universal SKILL.md specification. Once committed to .agents/skills/presentations/SKILL.md in your repository root, any supported tool will parse the manifest and trigger rules uniformly.
Q:How does Pitch Deck & Presentation Architect prevent context window overflow during long agent sessions?
Instead of passing raw visual XML or entire slide payloads across prompt turns, the skill instructs the agent to write and execute programmatic python-pptx builder scripts, passing only schema configs and test assertions through the context window.
Q:Does this skill require escalated root or container permissions?
No. The skill operates with a 98/100 security score, requiring only unprivileged local filesystem writes and Python package execution within workspace-scoped boundaries without elevated permissions.
Ready to equip your autonomous agent with Pitch Deck & Presentation Architect?
Copy the complete audited SKILL.md specification or download the 1-click zip bundle directly from our verified registry.