Influencer

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

assistant/CLAUDE.md

assistant/src/approvals/CLAUDE.md

assistant/src/notifications/CLAUDE.md

assistant/src/runtime/CLAUDE.md

gateway/CLAUDE.md

Audit 0.4.1 · source sha 19bc86343b63 · 14 Oct 2026Report an issue or request a re-scan

• Publishervellum-ai GitHub organizationOwner of the repository this skill is read from
• Source foldervellum-ai/vellum-assistant/skills/influencerThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict19bc86343b63A new audit runs when the source changes
• Repository statusNot archivedLast push 1 day ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Research influencers on Instagram, TikTok, and X/Twitter through your browser

---
name: influencer
description: Research influencers on Instagram, TikTok, and X/Twitter through your browser
compatibility: "Designed for Vellum personal assistants"
metadata:
  emoji: "🔍"
  vellum:
    category: "browsing"
    display-name: "Influencer Research"
    includes: ["vellum-browser-use"]
---

Use browser automation for collection and `host_bash` helper scripts for deterministic parsing, scoring, and comparison. All browser operations are executed through the `assistant browser` CLI, invoked via `host_bash`.

## Required tools

- `host_bash` for `assistant browser` CLI commands and helper scripts in `scripts/`.

## Step graph (state machine)

### Step 1: Route intent

Use deterministic routing when intent is unclear:

```bash
bun {baseDir}/scripts/influencer-intent.ts --request "<latest user request>" --has-candidates <true|false> --has-shortlist <true|false>
```

Use returned `step` to route to `discover`, `enrich_profile`, or `compare_shortlist`.

### Step 2: Discover candidates (`discover`)

#### Instagram

1. Navigate to keyword search/post surfaces.
2. Snapshot + extract:

```bash
assistant browser --session influencer --json snapshot
assistant browser --session influencer --json extract --include-links
```

3. Parse candidates:

```bash
bun {baseDir}/scripts/influencer-parse-candidates.ts --platform instagram --input-json '<json payload with extracted text/links>'
```

#### TikTok

1. Navigate to user search page for query.
2. Use `assistant browser --session influencer scroll` + `assistant browser --session influencer wait-for` to load additional candidates.
3. Extract and parse:

```bash
bun {baseDir}/scripts/influencer-parse-candidates.ts --platform tiktok --input-json '<json payload with extracted text>'
```

#### X/Twitter

1. Navigate to people search view (`f=user`).
2. Snapshot + extract:

```bash
assistant browser --session influencer --json snapshot
assistant browser --session influencer --json extract --include-links
```

3. Parse:

```bash
bun {baseDir}/scripts/influencer-parse-candidates.ts --platform twitter --input-json '<json payload with extracted text/links>'
```

### Step 3: Enrich profiles (`enrich_profile`)

For each selected candidate profile:

1. Navigate to profile URL.
2. Snapshot + extract profile metadata (bio, follower counts, verification indicators).
3. Score with criteria:

```bash
bun {baseDir}/scripts/influencer-score.ts --query "<user query>" --min-followers <n> --max-followers <n> --verified-only <true|false> --input-json '<json payload with profiles>'
```

4. If themes are missing, enrich using:

```bash
bun {baseDir}/scripts/influencer-theme-extract.ts --bio "<bio>" --query "<user query>"
```

### Step 4: Build shortlist (`compare_shortlist`)

Generate deterministic comparison output:

```bash
bun {baseDir}/scripts/influencer-compare.ts --limit <n> --input-json '<json payload with profiles and criteria>'
```

Present results grouped by platform with:

- Username / display name
- Followers (normalized)
- Verification status
- Theme highlights
- Profile URL

## Retry and fallback policy

- Retry budget: 3 attempts for each state-changing browser step.
- After any navigation or click that changes DOM, run fresh `assistant browser --session influencer --json snapshot`.
- If blocked by sign-in wall or challenge after retries, ask user to complete that step and resume from latest successful state.

## Platform notes

- Instagram search often surfaces posts/reels before profiles; use author-handle pivot logic.
- TikTok search can require scroll cycles to load profile cards.
- X/Twitter should use people-search surfaces to avoid irrelevant mixed-content feeds.

## Example helper payload shape

```json
{
  "phase": "discover",
  "context": { "platform": "instagram" },
  "extracted": {
    "text": "...",
    "links": ["https://www.instagram.com/example/"]
  },
  "userIntent": "find fitness creators"
}
```

Excerpt, first 3,924 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/vellum-ai/vellum-assistant
git -C vellum-assistant checkout 19bc86343b63

Works in

1. 2026-10-14
   ✓SAFE B 89 / 100

   first audit · source 19bc86343b63

Questions about Influencer

• github.com/vellum-ai/vellum-assistant/tree/main/skills/influence — the audited source at commit 19bc86343b63 and every finding aboveObserved · 14 Oct 2026

Spotted something wrong? Report a correction · Methodology