Spec: LinkedIn Post Writer

1. Current State

Pete has no LinkedIn post writing capability in his OpenClaw setup. He has been using a custom GPT with a detailed voice guide for drafting posts. His existing skills cover Twitter briefings, LinkedIn invites, and personal one-pagers, but nothing for content creation. No content idea capture system exists either.

2. Platform Capabilities

OpenClaw skills can bundle reference files (voice guides, style docs) that get loaded on demand. The skill trigger system uses the description field in frontmatter to match user intent. Skills can call web_fetch to read articles, and the agent can draft directly in chat. OpenClaw does NOT have native LinkedIn posting (no API integration). There is a community extension on ClawHub for LinkedIn posting via browser automation, but for drafting only, no integration is needed since the output is text Pete copies and pastes.

3. Community Patterns

A few community approaches exist: one ClawHub skill patches browser automation for direct LinkedIn posting with image uploads. Most people keep it simpler: a skill that drafts text in chat and the human posts it. The GPT Custom Instructions pattern Pete was using is common and translates directly to a skill reference file. No one is doing anything more sophisticated for voice-matched content drafting specifically.

4. Options

Option A: Draft-only skill. A skill with a voice guide reference file. Triggered by "write a LinkedIn post." Drafts in chat. Pete copies to LinkedIn manually.

• Complexity: Low
• Token cost: ~2k per draft (loading voice guide + drafting)
• Reliability: High (no external dependencies)
• Maintenance: Minimal

Option B: Draft + auto-post via browser relay. Same as A but after approval, posts directly to LinkedIn via browser automation.

• Complexity: High (browser relay needed, LinkedIn DOM changes break it)
• Token cost: Same plus browser session cost
• Reliability: Fragile (LinkedIn DOM changes frequently, anti-bot detection)
• Maintenance: High (constant fixing when LinkedIn updates)

Option C: Draft + LinkedIn API. Use LinkedIn's official API for posting.

• Complexity: Medium (OAuth setup, app registration)
• Token cost: Same as A
• Reliability: Medium (API access is restrictive, requires LinkedIn app approval)
• Maintenance: Medium (API versioning)

5. Recommendation

Option A. Draft-only. The value is in the voice matching and idea distillation, not the posting mechanism. Copy-paste to LinkedIn takes 10 seconds. Browser automation and API integration add fragility for almost no time savings. We can add posting later if the volume justifies it.

6. Security Considerations

• No external API calls or credentials needed (draft-only)
• Voice guide contains writing style preferences, not sensitive data
• If articles are fetched for inspiration, standard web_fetch rules apply (http/https only)
• No risk of accidental posting since there is no LinkedIn integration

7. Implementation Scope

David builds:

• skills/linkedin-post-writer/SKILL.md with frontmatter, workflow steps, input types, quality checks
• skills/linkedin-post-writer/references/voice-guide.md adapted from Pete's GPT prompt (Pete provides the source content, David writes the file)
• X/Twitter adaptation section for future cross-posting format

8. Validation Criteria

• Give it a raw idea and verify the output follows voice rules (no em dashes, no bullets, no corporate filler, short paragraphs, concrete specifics)
• Give it an article link and verify it extracts Pete's angle rather than summarizing the article
• Give it an existing draft and verify it tightens without losing Pete's voice
• Verify reflection step happens before drafting (2-4 sentence insight distillation)