li-audit

What already went out, judged honestly. Ranks by the metrics that matter, finds the pattern, and feeds the result straight into next week's li-plan.

Before you audit

1. Read the history: ~/.agents/linkedin/log.md (then the Claude/Copilot fallbacks), which li-post appends to on every approved post. If the log is empty or missing, ask the user to paste their recent posts with stats.
2. Get the numbers per post: impressions, reactions, comments, reposts, and ideally profile follows or DMs attributed. Work with what exists; mark missing metrics as unknown rather than inferring them.

The metrics that matter

Compute per post, and rank by these, not by raw impressions:

• Engagement rate = (reactions + comments + reposts) / impressions. The headline metric. Above 3% is strong for most niches.
• Reach multiple = impressions / follower count. Above 3x means the algorithm carried it beyond the audience.
• Comment quality. Count substantive comments separately from congratulations. Ten "great post"s lose to two real questions.
• Conversion. Profile visits, follows, DMs, leads. Self-reported; take the user's word.

Show the ranking table with all four columns. Then say what the table means in two sentences. Tables do not interpret themselves.

The diagnosis

Find the pattern across the top 3 and bottom 3:

• Hook family: which of the 21 formulas won? (Cross-check with hooks.json.)
• Format: text vs carousel vs repurposed.
• Pillar: which content pillar is carrying the account?
• Topic: what subject, specifically, not "AI content".
• Length and structure: did the winners share a shape?

Then the prescription, as inputs to li-plan:

AUDIT
posts reviewed: 12 (4 weeks)
top:     "The proposal story" - 4.1% engagement, 6.2x reach (hook #17, story)
bottom:  "AI trends 2026" - 0.8% engagement, 1.1x reach (generic take)
pattern: stories with numbers travel; generic takes die
prescribe: 2 story posts next week, kill the trends pillar, test carousel
           for the framework post

Rules

• Judge the post, not the person. "This post failed" is useful; "your writing is weak" is not.
• Small samples lie. Under 8 posts, present findings as hypotheses with "test:" labels, not conclusions.
• Do not recommend posting more as the fix for low engagement. The fix is almost always sharper ideas or better hooks, not volume.
• If a post the user loved flopped, say so kindly and say why. The audit is the one skill allowed to disagree with the user's taste.
• Write the prescription into the next li-plan run. An audit nobody acts on is entertainment.