Weekly Roundup

Weekly Roundup Skill

Purpose

Automate the Tuesday LinkedIn roundup for the AI Capability Reference. This skill handles four jobs:

1. Read scan results and identify what changed this week.
2. Select the right post template based on the week's changes.
3. Draft a Post Protocol-compliant LinkedIn post.
4. Log publication to LocalBrain after posting.

Dependencies

This skill depends on two reference files. Read both before drafting.

• references/post-protocol-extract.md -- the subset of the Post Protocol that governs this post slot (lane, packaging order, compliance rules, measurement).
• references/roundup-templates.md -- the three roundup templates with selection criteria and a worked example.

The shared LinkedIn voice profile lives in the linkedin-post skill. If that skill is available, defer to its voice reference and compliance check. If not, references/post-protocol-extract.md contains the minimum rules needed to draft compliantly.

Workflow

Step 1: Gather changes

Identify what changed since the last roundup. Sources, in priority order:

1. GitHub issues closed since last Tuesday (label: scanner, verification, or manual). Use gh issue list --state closed --since <date> if available, or ask the user.
2. Git log of commits to data/ since last Tuesday.
3. User's verbal summary of what they updated.

If no structured source is available, ask the user: "What changed in the capability reference this week?" Do not draft without changes.

If nothing changed, say so. A week with no changes means no roundup post. Do not invent content.

Step 2: Select template

Read references/roundup-templates.md. Three templates exist:

• Changes-first (default): Use when there are 2+ concrete platform changes to report. Leads with specifics.
• Pattern-focused: Use when the week's changes reveal a trend or pattern across platforms (e.g., multiple providers expanding free tiers, or a wave of GA promotions). Leads with the tension.
• Shortest-viable: Use when the user signals low energy, when changes are minor, or when there are fewer than 2 notable changes. Keeps the streak alive.

Selection rules:

• 3+ notable changes across 2+ platforms --> changes-first
• 2+ changes that share a pattern --> pattern-focused
• 1 change or minor updates only --> shortest-viable
• User says "keep it short" or similar --> shortest-viable

Present the selection and rationale before drafting. The user may override.

Step 3: Draft the post

Follow these rules in order:

1. Read the linkedin-post skill's voice reference if available. Otherwise read references/post-protocol-extract.md.

2. Apply the selected template from references/roundup-templates.md.

3. The post must fit the Bridge lane (AI onramps for humans, practical enablement).

4. Apply packaging order: a. Expensive failure (what goes wrong if you miss this) b. Human behavior under pressure (how trainers/buyers get burned) c. Instrument (the tracker as the verification tool) d. Smallest fix (bookmark it, check before your next session)

   Not every element needs to be explicit. The packaging order is a bias, not a checklist. The changes-first template naturally foregrounds the instrument. The pattern-focused template naturally foregrounds the expensive failure.

5. The capability reference link goes at the end of the post, inline. Never in comments.

6. Hashtags: exactly 2. Default: #AIStrategy #FutureOfWork. Swap one if the week's changes strongly fit a different lane (e.g., #AIGovernance for policy-relevant changes).

7. CTA: one only. For changes-first and shortest-viable, the link is the CTA. For pattern-focused, a precise question is acceptable as an alternative.

Present 2 draft options using the selected template but with different hook angles. The user picks one.

Step 4: Compliance check

Run this check on the selected draft before delivering. If the linkedin-post skill is available, use its full compliance check instead.

Checks (fail if any violated):

• Plaintext only (no markdown emphasis)
• No em dashes
• No emoji
• No bold, no italics
• Paragraphs scannable (1-3 sentences each)
• First 2 lines work as standalone preview above the fold
• One primary signal
• No hype or guru language
• No influencer patterns ("Here are 7 things...")
• Link at end of post, inline
• Hashtags at end, 2 only
• Character count 900-1,300
• CTA is not needy
• Lane is Bridge
• Packaging order is present as bias

Report: "Compliance: pass" or list violations and fix them before delivering.

Step 5: Log publication

After the user confirms the post has been published, log it to LocalBrain using the log_published tool:

• title: "AI Capability Reference Weekly Roundup - [date]"
• channel: linkedin
• company: paice
• content_type: promo
• description: One-line summary of what the post covered

If LocalBrain is not available, append to the daily note instead.

If neither is available, remind the user to log it manually.

Quick Reference