Instagram audit

Turn the account's own numbers into a short list of claims with evidence attached, each with a confidence level and one action. Raw totals mostly measure follower count. Compare each post against the account's own median instead.

1. Pull the data

Run from the instagram skill directory (../instagram):

npx -y bun scripts/main.ts audit [username] --limit 30 --tz <Area/City> --json > "<project-dir>/instagram-audit.json"
npx -y bun scripts/main.ts profile [username] --json > "<project-dir>/instagram-profile.json"   # default: own account

• <project-dir> is the user's working folder (absolute path), never the plugin folder.
• Omit username for the signed-in account. Ask which timezone the audience lives in; the default is the machine's.
• On rate_limited, checkpoint or not_logged_in, stop and report the hint. Do not retry.
• Check sample.source. "embedded, max 12" means only 12 posts could be read, so every pattern is weak. Say so.

2. Read what the audit returns

Never present a bucket with n < 3 as a finding. Instagram hides like counts on some posts (null). Those are excluded, not counted as zero.

3. Find the pattern, in this order

1. Outliers vs underperformers. Put the top 5 and bottom 5 side by side. Compare the hook (the caption's first line, or the on-screen hook if the user describes it), format, length and theme. Score the hooks with npx -y bun ../instagram-content/scripts/hook-score.ts --hook "<line>" when useful.
2. Format. Reels vs carousels vs photos, by median engagement rate and medianPlays.
3. Caption length and hashtags. Usually minor. Report them only when the gap is large and n is solid.
4. Cadence. A gap longer than 14 days usually explains a "reach drop" better than anything else.
5. Day and hour, last. They are almost never the cause. Mention them only when nothing else shows up and the buckets have n of 3 or more.

State each finding as a claim with its evidence (numbers, n, example posts) and its confidence (high, medium or low). Thirty posts can show a pattern. Six cannot, so say that instead of inventing one.

4. Separate hook problems from profile problems

• Low plays or engagement across the board means a hook or content problem. Hand off to instagram-content.
• Posts that travel while the account does not grow means a profile problem. The web session cannot measure follows per post, so ask for Insights (non-follower reach, follows) to confirm.

5. Score the profile

Score instagram-profile.json (name field, bio, link, category, business account) against references/profile-rubric.json (12 items, 100 points). Items the data cannot show (highlights, pinned posts, photo, story ring) are scored from a screenshot or the user's description, or marked "not checked". Score strictly. Most first-pass profiles land in the 30s and 40s. Write replacement copy for the 3 lowest items (name field 30 chars, bio 150 chars). Profile edits happen in the app. This plugin does not change profiles.

6. Output

INSTAGRAM AUDIT · @handle · 30 posts (2026-08-01 to 2026-11-01, America/Bogota) · source: REST
Baseline: median 412 engagements, 2.1% rate, reels median 9.8k plays
Profile score: 46/100 (name field 2/12, bio line 4/12, ...)

FINDINGS
1. Carousels outperform reels on this account: median rate 3.4% (n=9) vs 1.6% (n=14). Confidence: medium.
   Action: two carousels a week for the next 4 weeks, then re-run this audit.
2. ...

NOT VISIBLE FROM THE WEB SESSION
reach, saves, sends, watch time, follows per post -> paste Insights to sharpen findings 1 and 3.

Keep it to 3 to 5 findings, one action each, plus a re-audit date. See references/metrics.md for definitions and the metrics that need Insights.

Additional resources

• references/profile-rubric.json: 12-item, 100-point profile rubric.
• references/metrics.md: metric definitions, benchmarks, and what to ask for from Insights.