Prompt Of The Week

Prompt-of-the-Week PPTX Skill

Think of this skill as a mail merge meets slide design: the GitHub repo is the data source, and the Prompt-K.pptx template is the envelope. Your job is to pour the right data into the right slots without disturbing the surrounding design.

Quick Overview

Author Photo Convention (IMPORTANT)

The skill ships with an author/ folder that always contains exactly one image — the photo of the current week's author. Always use that file as the photo for the slide. Do not attempt to download avatars from GitHub or the web — this sandbox has no outbound internet access for binary fetches.

SKILL_DIR="/mnt/user-config/.claude/skills/prompt-of-the-week"
AUTHOR_IMG=$(ls "$SKILL_DIR/author/" | head -1)   # exactly one file expected
AUTHOR_PATH="$SKILL_DIR/author/$AUTHOR_IMG"

If the folder is empty or missing, ask the user to drop the week's author photo into prompt-of-the-week/author/ before continuing. The user maintains this folder — replacing the image each week is part of their workflow.

To swap it into the slide (the visible photo on the right side is Picture 30 / id=31 / rId5 → image5.png, NOT the off-slide Image 1 / id=3 / rId3):

1. Convert the author image to PNG and overwrite working/unpacked/ppt/media/image5.png. Keeping the filename means no rels or [Content_Types].xml edits are needed:

   from PIL import Image
   Image.open(AUTHOR_PATH).save('working/unpacked/ppt/media/image5.png', 'PNG')
   

2. Do NOT touch Image 1 (id=3, rId3) — that shape sits off-slide and is not visible. The skill's earlier instructions referencing rId3 were wrong; rId5 is the correct target.
3. The Picture 30 shape already has photo effects (sharpen, brightness/contrast) and a 3D bevel applied — these stay as-is and apply to the new image automatically. Square (1:1) author images render best.

Step 1 — Fetch Data from GitHub

The user will provide a URL like:

https://github.com/pnp/copilot-prompts/tree/main/samples/agent-instructions/creator-agent

Convert it to raw content URLs and fetch both files:

# Convert tree URL → raw URL pattern
# https://github.com/{owner}/{repo}/tree/{branch}/{path}
# → https://raw.githubusercontent.com/{owner}/{repo}/{branch}/{path}/README.md
# → https://raw.githubusercontent.com/{owner}/{repo}/{branch}/{path}/assets/sample.json

SAMPLE_PATH="samples/agent-instructions/creator-agent"  # extracted from URL
README_URL="https://raw.githubusercontent.com/pnp/copilot-prompts/main/${SAMPLE_PATH}/README.md"
META_URL="https://raw.githubusercontent.com/pnp/copilot-prompts/main/${SAMPLE_PATH}/assets/sample.json"

curl -sSL "$README_URL" -o /home/claude/sample_readme.md
curl -sSL "$META_URL" -o /home/claude/sample_meta.json 2>/dev/null || echo "{}" > /home/claude/sample_meta.json

If curl fails (network restrictions), use the web_fetch tool directly on the raw URLs.

Step 2 — Extract the Required Fields

Parse the fetched content with Python. Every field maps to exactly one shape in the template.

import json, re

readme = open('/home/claude/sample_readme.md').read()
meta   = json.load(open('/home/claude/sample_meta.json')) if open('/home/claude/sample_meta.json').read().strip() != '{}' else {}
sample_parts = SAMPLE_PATH.strip('/').split('/')
sample_folder = sample_parts[-2] if len(sample_parts) >= 2 else ''

# Top tile varies by source folder.
if sample_folder == 'agent-instructions':
    title_top = 'System Prompt of the Week'
elif sample_folder == 'skills':
    title_top = 'Skill of the Week'
else:
    title_top = 'Prompt of the Week'

# ── Field 1: Agent / Prompt Title ──────────────────────────────────────────
# Prefer: first H1 heading in README; fallback to meta['title']
title_match = re.search(r'^#\s+(.+)', readme, re.MULTILINE)
agent_title = title_match.group(1).strip() if title_match else meta.get('title', 'Unknown Agent')

# ── Field 2: The Prompt (system prompt body) ────────────────────────────────
# Look for a code block (```markdown / ``` / ```text) that contains the prompt,
# OR a section headed "## Prompt" / "## Instructions" / "## System Prompt"
prompt_body = ''
block = re.search(r'(?:##\s+(?:Prompt|Instructions?|System Prompt)[^\n]*\n)(.*?)(?=\n##|\Z)', readme, re.DOTALL)
if block:
    prompt_body = block.group(1).strip()
    # Strip fenced code delimiters if present
    prompt_body = re.sub(r'^```[a-z]*\n?', '', prompt_body, flags=re.MULTILINE).strip('`\n ')

# ── Field 3: Author name ────────────────────────────────────────────────────
# Check meta['authors'][0]['name'], then README "Author:" line
author = ''
if meta.get('authors'):
    author = meta['authors'][0].get('name', '')
if not author:
    auth_match = re.search(r'[Aa]uthor[:\s]+([^\n]+)', readme)
    if auth_match:
        author = auth_match.group(1).strip()

# ── Field 4: Compatibility / tags summary (optional, shown in footer area) ──
tags = ', '.join(meta.get('tags', []))

# ── Field 5: Sample folder name (used as slug / emoji label) ────────────────
sample_name = SAMPLE_PATH.split('/')[-1].replace('-', ' ').title()

print(f"Top tile: {title_top}")
print(f"Title:   {agent_title}")
print(f"Author:  {author}")
print(f"Tags:    {tags}")
print(f"Prompt preview: {prompt_body[:120]}...")

Use these folder rules when setting the top tile in the slide:

Analogy: The slide fields are like variables in a template literal — ${title}, ${author}, etc. The XML editing in Step 3 is the interpolation engine.

Step 3 — Edit the Template

See references/xml-field-map.md for the exact XML shape names and replacement strategy.

# 1. Copy template to working location
cp /path/to/skill/references/Prompt-K.pptx /home/claude/output.pptx

# 2. Unpack
python /mnt/skills/public/pptx/scripts/office/unpack.py /home/claude/output.pptx /home/claude/unpacked/

# 3. Edit slide1.xml  ← all content lives on the single slide
#    (Use str_replace tool — one replacement per shape)
#    Set the top title shape to title_top based on SAMPLE_PATH:
#    agent-instructions => "System Prompt of the Week"
#    prompts => "Prompt of the Week"
#    skills => "Skill of the Week"

# 4. Clean & repack
python /mnt/skills/public/pptx/scripts/clean.py /home/claude/unpacked/
python /mnt/skills/public/pptx/scripts/office/pack.py /home/claude/unpacked/ /home/claude/output.pptx \
       --original /path/to/skill/references/Prompt-K.pptx

When editing XML text nodes, always preserve surrounding <a:r> run tags and their <a:rPr> formatting. Only replace the inner <a:t> content. Changing run properties will break fonts and colours.

Step 4 — QA & Present

# Convert to image for visual check
python /mnt/skills/public/pptx/scripts/office/soffice.py --headless --convert-to pdf /home/claude/output.pptx
pdftoppm -jpeg -r 150 /home/claude/output.pdf /home/claude/slide
ls /home/claude/slide*.jpg

Visually verify (use the view tool on the jpg):

• Top tile text matches the sample folder (agent-instructions, prompts, or skills)
• Title text matches the sample name
• Prompt body is not cut off or overflowing the dashed border box
• Author name is visible in the right-side name badge
• The author photo on the right matches the image from the skill's author/ folder
• No leftover placeholder text ("The PROMPT…", "Creator Agent", "Cyprien Maloeuvre")

Then copy to outputs and present:

cp /home/claude/output.pptx /mnt/user-data/outputs/PromptOfTheWeek.pptx

Reference Files

Read references/xml-field-map.md before you open slide1.xml.