# Social Audit

> >

- **Type:** Skill
- **Install:** `agentstack add skill-inklate-social-skills-social-audit`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [inklate](https://agentstack.voostack.com/s/inklate)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [inklate](https://github.com/inklate)
- **Source:** https://github.com/inklate/social-skills/tree/main/skills/social-audit
- **Website:** https://inklate.com

## Install

```sh
agentstack add skill-inklate-social-skills-social-audit
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

Audit the profile and content history the user gives you against their stated goals, and hand back a prioritized gap list with a 3-move action plan.

## Context

Read `social-context.md` at the project root (also check `.agents/social-context.md`) for the `## Goals`, `## Pillars`, `## Audience`, and `## Cadence` sections — the audit measures reality against these, so they matter more here than in any other task. If it's missing, offer to run the `social-context` skill first, but don't block — ask 2–3 quick inline questions and proceed:

- What's the goal — leads, audience growth, hiring, authority?
- What pillars did you _intend_ to post about?
- What cadence did you _intend_ to keep?

## Workflow

1. **Collect the evidence.** Ask the user to paste what they have:
   - recent posts, text plus dates — ideally the last 20–30;
   - the bio/profile text, the link, and what's pinned;
   - any numbers they can report: follower trend, typical impressions/likes, their best and worst post.
     Work only from what's pasted. State clearly up front what you cannot assess without more — e.g. "without impression data I can score your hooks but not your reach; paste your top and bottom 5 posts by impressions if you have them." Never fill an evidence gap with a guess.
2. **Score posting consistency.** From the post dates, compute actual posts-per-week and compare to the intended cadence from context. Flag gaps longer than 2× the intended interval, and flag clustering — five posts in three days, then two weeks of silence, is not a 2.5/week cadence. Check this dimension first: consistency gaps explain "flat" more often than content quality does, and every other score is noisy until cadence is stable.
3. **Score pillar balance.** Tag each pasted post with its pillar, or "off-pillar" if it fits none. Report the actual distribution against the intended pillars, with counts. Be specific and honest: "your last 12 posts are all pillar #1" — never soften it to "consider diversifying your content". Both failure modes are findings: single-pillar collapse and off-pillar drift.
4. **Score hook quality.** Take the first line of each of the last N posts (N = what they pasted, cap 20) and score each 0–2:
   - **0** — label or throat-clearing: "Some thoughts on hiring." / "I've been meaning to write this."
   - **1** — clear topic, no tension: "How we run our onboarding."
   - **2** — specific claim, tension, or curiosity gap that survives truncation: "Our onboarding had a 60% drop-off at step 2. One email fixed it."
     Report the average, then quote the worst three verbatim with rewrites — the rewrites teach more than the scores do. Then check fit: flag any hook aimed at the wrong reader for the `## Audience` in context — a strong hook for the wrong audience still misses.
5. **Score format mix.** Count the formats used — text post, thread, carousel/document, image, video/Reel — against what the platform rewards and what the goal needs. Two failure modes, both findings: one-format monoculture, and format-chasing where nothing is tried twice so nothing can be learned.
6. **Read the engagement pattern — from their numbers only.** If the user reported numbers, look for the shape:
   - which pillar and format overlap their best posts;
   - whether engagement concentrates in a few spikes or sits flat across everything;
   - replies-vs-likes ratio if known — replies signal resonance, likes-only signals passable-but-skippable.
     If they reported nothing, write "unknowable from pasted data" for this dimension and move on. Do not infer engagement from post text.
7. **Audit the profile itself.** A growth problem is often a conversion problem — people arrive from a good post and bounce off a vague bio. Check line by line:
   - first line of the bio: does it name the audience from `## Audience` and the offer from `## Positioning` — who it's for and what they get — or is it a job title and three emoji?
   - credibility: any evidence — numbers, names, track record?
   - the link: present, working destination described, matched to the goal?
   - pinned post: their best converter, or just their newest?
   - profile photo and banner: present and legible at feed size, as described?
8. **Diagnose against the goal.** Connect the findings to the stated goal, not to generic best practice: flat followers + solid consistency + weak hooks is a top-of-funnel problem; good reach + no leads is a bio/CTA problem; erratic cadence means nothing else is measurable yet — fix that before optimizing anything. Name the single biggest lever explicitly, in one sentence.
9. **Build the gap list.** Every finding as one line: severity (high/med/low), the evidence (a quote, count, or date range from their own data), and the fix. Order by expected impact on _their_ goal, not by ease of fixing.
10. **Write the 3-move action plan.** Exactly three moves, biggest lever first, each concrete enough to start this week — "rewrite your bio's first line to name the audience; here's a draft", not "improve your profile". Wherever a move has a draft or example, include it inline.

## Quality bar

| Dimension      | Evidence required               | Verdict style                              |
| -------------- | ------------------------------- | ------------------------------------------ |
| Consistency    | Post dates vs stated cadence    | Actual number vs intended number           |
| Pillar balance | Per-post pillar tags            | Distribution counts, drift named           |
| Hooks          | First lines, scored 0–2         | Average, plus worst 3 quoted with rewrites |
| Format mix     | Format counts                   | Monoculture or scattershot, named          |
| Engagement     | User-reported numbers only      | Pattern, or "unknowable from pasted data"  |
| Profile        | Pasted bio / link / pinned post | Line-by-line, with rewrite drafts          |

Hard rules:

- Every claim cites the user's own data — a quote, a count, a date range. If you can't point at evidence, it isn't a finding.
- No invented benchmarks: "posts with images get 2.3× engagement" is not something you know from pasted data.
- Unknowns are stated as unknowns, never papered over with plausible-sounding filler.
- "Diversify your content"-grade advice is banned — name the posts, name the gap, show the fix.
- Severity reflects impact on _their_ stated goal, not general social-media hygiene.
- The 3-move plan is exactly three moves; a ten-item to-do list is an audit that dodged prioritization.
- Every dimension in the table above gets a verdict or an explicit "couldn't assess — need X"; a silently dropped dimension reads as a pass.

## Deliverable

The audit report: a one-paragraph verdict naming the single biggest lever, the dimension-by-dimension findings with quoted evidence, a "couldn't assess" list naming exactly what data would unlock each missing item, the prioritized gap list, and the 3-move action plan with inline drafts wherever a move includes one. This is analysis and a plan for the user to act on themselves — hand it back and stop.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [inklate](https://github.com/inklate)
- **Source:** [inklate/social-skills](https://github.com/inklate/social-skills)
- **License:** MIT
- **Homepage:** https://inklate.com

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-inklate-social-skills-social-audit
- Seller: https://agentstack.voostack.com/s/inklate
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.