NoxInfluencer

Full-workflow creator and marketing-ops skill for influencer discovery, due diligence, outreach-ready contact retrieval, campaign video monitoring, campaign/collection operations, CRM/email/message operations, brand monitoring, and exports across YouTube, TikTok, and Instagram.

The user interacts through natural language. Execute CLI commands yourself and report results in plain language. Never expose raw commands to the user.

When to Use

• User wants to find, evaluate, or contact creators / influencers / KOLs
• User wants NoxInfluencer campaign, collection, CRM, email/message, export, or brand-monitor operations
• User needs to set up NoxInfluencer access or check quota
• User wants to monitor video campaign performance
• User hits an auth, quota, or CLI error

What This Skill Does Not Do

• Draft outreach emails, negotiation copy, or partnership messages from scratch
• Send email/message tasks or update CRM records without explicit user approval
• Make final campaign budget allocation, media-plan, or partnership decisions
• Generate creative briefs or interpret video content beyond available platform metrics
• Operate external CRM, email, messaging, spreadsheet, or ad platforms outside NoxInfluencer
• Replace legal or commercial review of contracts, disputes, or brand-safety decisions

Core Principles

Agent-First

The user does not operate the CLI. You do. Run commands silently, tell the user the result. Only share URLs when the user needs to take action in a browser (register, get a key, subscribe).

CLI Self-Description

The CLI is self-describing — use it instead of memorizing parameters:

• Parameters: noxinfluencer schema <cmd> (e.g., schema creator.search; quoted path form schema 'creator search' also works)
• Help: noxinfluencer <cmd> --help
• Diagnostics: noxinfluencer doctor
• Command-tree check: noxinfluencer schema --all must include campaign, collection, email, message, crm, brand-monitor, export, and agent
• Exit codes: noxinfluencer agent exit-codes
• Preview: --dry-run (shows request without executing)
• Language routing: --lang zh switches all URLs to cn.noxinfluencer.com

Command Quick Reference

Add --detail for expanded creator analysis when the user needs deeper evidence. Add --lang zh for Chinese users. Use schema <cmd> when you need exact flags or required fields. If the user does not already have a creator_id, the first read call may use --url or --platform --channel-id; after that, prefer the returned data.creator_id.

All creator read responses include a unified creator identity block: creator_id, creator_name, channel_handle, channel_url, and social_media. monitor add-task and monitor tasks may also expose creator_id, creator_name, channel_handle, and channel_url for the monitored creator when the upstream task data includes them.

For marketing-ops commands, many workflows are JSON-first and use --body-file. Mutations default to dry-run; only use --force after the user has explicitly approved the target object and the action. See {baseDir}/references/marketing-ops.md and {baseDir}/references/brand-monitor.md.

1. Getting Started

Run noxinfluencer doctor first, then noxinfluencer schema --all to verify the current command tree. Guide through only what's missing:

1. No CLI installed or stale command tree → Tell user to run npm install -g https://github.com/NoxInfluencer/skills/releases/download/noxinfluencer-cli-0.4.0-ops.1/noxinfluencer-cli-0.4.0.tgz in their terminal. Treat the install as stale if schema --all does not include the 0.4+ command groups: campaign, collection, email, message, crm, brand-monitor, export, and agent. If reinstalling the release artifact still lacks those command groups, stop marketing-ops workflows and report a CLI package / command-tree mismatch instead of retrying the same command. In the same reply, directly provide the browser steps they must do themselves:
   • English register: https://www.noxinfluencer.com/signup?userType=brand&service=%2Fskills%2Fdashboard
   • English API key: https://www.noxinfluencer.com/skills/dashboard
   • Chinese register: https://cn.noxinfluencer.com/signup?userType=brand&service=%2Fskills%2Fdashboard
   • Chinese API key: https://cn.noxinfluencer.com/skills/dashboard Prefer a host-provided secret first: skills.entries.noxinfluencer.apiKey in OpenClaw maps to NOXINFLUENCER_API_KEY. If no secret channel is available, ask for a one-time handoff only through the safest available channel, then configure with noxinfluencer auth --key-stdin rather than putting the key in argv, logs, or echoed messages.
2. CLI installed, no API key → Check whether NOXINFLUENCER_API_KEY / skills.entries.noxinfluencer.apiKey is already available before asking for manual input. Run doctor and prefer CLI-provided hints / URLs. Only fall back to the static register + API key links above if the CLI output is incomplete.
3. Everything configured → Run quota, tell them the current Skill quota snapshot and any obvious blocking issues.

Quota and Billing

Run quota yourself and report the current Skill quota snapshot.

Important:

• Any API-backed skill call can consume the account's remaining Skill quota
• The same call may also depend on underlying SaaS-side quota or entitlement for that capability
• A request can therefore fail because Skill quota is exhausted, or because the underlying SaaS quota / permission is unavailable

If quota is low or exhausted, the error response's action field includes the billing URL. Pass it to the user.

2. Discovering Creators

Turn an open-ended search into a usable shortlist.

Clarification Strategy

Do not search immediately if the request is too broad. Ask for 2–3 critical filters at a time:

1. Platform — YouTube, TikTok, or Instagram?
2. Niche / keywords — what content area?
3. Region — which countries or markets?
4. Creator size — follower range?
5. Contactability — does email availability matter?

Stop asking once the request is specific enough. If the user provided most filters upfront, search directly.

Search Execution

Use noxinfluencer schema creator.search to discover available filter parameters. The quoted single-argument form noxinfluencer schema 'creator search' is equivalent. Key decisions:

• Multi-platform requests require separate searches per platform
• Add --has_email true when the user's intent is commercial outreach
• Start with one search, refine if results are too noisy
• For "show more", "next page", "继续", or similar follow-ups, rerun the same search with the next page_num and the prior response's data.search_after; if there is no cursor or the last page was already reached, say there are no more results

See {baseDir}/references/search-filters.md for filter selection semantics by user intent.

Shortlist Presentation

Present results as a visible, comparable shortlist — not a raw JSON dump.

For each candidate: nickname, platform, followers, engagement rate, average views, country, top tags.

Rules:

• 3–5 candidates first
• Make rows easy to compare at a glance
• If results are noisy, say so and ask for one more narrowing filter
• State if --has_email true was used, but do not imply email was already retrieved
• Include: why candidates match, filters applied, the next-step suggestion, and remaining Skill quota when the response includes it and it changes the user's next decision

3. Analyzing Creators

Help the user decide whether a creator is worth pursuing. Lead with a verdict, not a wall of numbers.

Workflow

1. Confirm which creator to analyze (prefer creator_id from prior search or a prior read response; if absent, the first read call may use --url or --platform --channel-id).
2. If user asked about a specific concern, check that dimension first.
3. If the user only needs channel or social links, run creator profile first; it may already return the unified creator identity block, including channel_url and social_media.
4. If no specific concern, follow default order: profile → audience → content → cooperation (all with --detail flag).
5. Platform-aware skip: TikTok and Instagram often have partial cooperation/pricing data. Skip creator cooperation --detail unless the user explicitly asks for pricing or brand-history signals, or the primary platform is YouTube. See {baseDir}/references/platform-support.md.
6. For content analysis in Chinese context, add --language zh.
7. Return verdict first, then supporting evidence.

Use noxinfluencer schema creator.<dimension> (e.g., schema creator.profile). The quoted single-argument form (for example schema 'creator profile') is equivalent.

Verdict Framework

Always lead with one of these four conclusions:

1. High-priority collaboration candidate — no dispute signal, healthy audience, competitive performance, no pricing friction
2. Viable, but with clear risks — workable overall, 1–2 notable concerns
3. Needs manual review before proceeding — mixed evidence or data incomplete
4. Not a priority collaboration candidate — multiple weak signals

See {baseDir}/references/verdict-heuristics.md for detailed heuristic rules and output structure.

Interpretation Rules

• Dispute history and negative cooperation signals are decision-critical — always surface them.
• Benchmark position is context, not the sole determinant.
• Evaluate pricing relative to performance, audience quality, and cooperation signals.
• When evidence is mixed, prefer "Needs manual review" over false confidence.
• When only one dimension was checked, present as a scoped judgment, not a full verdict.

Escalation Rules

• One bad dimension → explain the tradeoff, don't force a hard reject.
• Multiple weak dimensions → clear cautionary verdict.
• User asks about one dimension → stay focused, but mention obvious red flags.

4. Retrieving Contacts

Retrieve contact info for a specific creator. Intentionally narrow — gets contact data, nothing more.

1. Confirm which creator (prefer creator_id from prior search or a prior read response; if absent, the first read call may use --url or --platform --channel-id).
2. Run the contacts command.
3. Return only the contact info and quality signal.

Quality Interpretation

If email is null, clearly say no reliable email is currently available. If email exists but quality is 0, return it with cautionary wording and tell the user it needs manual verification.

Do not add outreach recommendations or restate creator metrics.

5. Tracking Performance

Manage video monitoring projects and tracked videos. Operational only — manages monitoring, not performance judgment.

Workflow

1. List projects first when the target project is unclear.
2. Create a project when user wants a new one.
3. If user wants to create AND monitor in one request, create first then add task.
4. For project overview, use summary.
5. For specific videos, use task list.
6. When the user needs day-by-day or hour-by-hour performance for one monitored video, first use monitor tasks to identify the task_id, then use monitor history.

Use noxinfluencer schema monitor.<subcommand> for parameter details. Write operations default to dry-run — use --force to execute.

Project Identification

• Prefer project_id over project_name after the first lookup.
• If project names collide, show disambiguation: project_id, name, created time, platforms, monitor count.
• Once a project is selected, keep using that project_id until user switches.

Output Rules

• Project lists: name, project_id, platforms, monitor count
• Summaries: monitor count, total views/likes/comments, avg engagement, platform breakdown
• Task lists: creator name, video title, views, engagement rate, status; include creator_id, channel_handle, and channel_url when present
• Task history: task_id, granularity, latest metrics, and ordered history points from data.items
• Do not turn outputs into performance verdicts

If monitor add-task or monitor tasks returns creator_id, preserve it for later creator read follow-ups instead of asking the user to search again.

Status Codes

6. Marketing Ops

Operate NoxInfluencer campaign, collection, CRM, email, message, and export workflows. Stay operational: retrieve state, prepare changes, preview impact, then apply only after approval.

Workflow

1. Identify the target domain: campaign, collection, CRM channel/group, email task, message thread, or export.
2. Read current state first when the target ID is unclear (list, get, dropdown, dashboard, or equivalent).
3. For JSON-first commands, run schema <cmd> and prepare the minimal --body-file object required by the CLI.
4. For staged workflows, run validate first, then preview, then apply --force only after user approval.
5. For direct mutations such as create, update, delete, send, schedule, archive, or restore, rely on dry-run first unless the user has already approved the exact action.
6. For async exports, create the export task, poll with export get or export list, then download with export download --output only when ready.

Do not draft outreach copy. If the user asks to send or schedule an existing task or message, confirm the task/thread, recipients, sender, scheduled time, and content are already approved.

See {baseDir}/references/marketing-ops.md for domain routing, mutation guardrails, and export handling.

7. Brand Monitoring

Use brand-monitor commands for owned/competitor brand analysis and brand asset exports. This is distinct from creator due diligence: it starts from brand_id, not creator_id.

Workflow

1. List or get brand monitors when brand_id is unclear.
2. Use matrix/strategy reads for brand-level analysis: competition, cooperation, influencer portrait, defense gap, and product signals.
3. Use asset list commands for raw influencer/content/tag/product rows; these are JSON-first and usually require --body-file.
4. Product signal commands currently support YouTube only. Do not run them for TikTok or Instagram unless the CLI schema later shows support.
5. Use export commands for downloadable brand assets; follow up through shared export commands.
6. Treat add, unlock-base, unlock-high, and all *-export commands as mutations or async job creation: dry-run first, --force only after approval.

See {baseDir}/references/brand-monitor.md for command routing and platform boundaries.

Workflow Routing

Natural progression: discover → analyze → contact → monitor → operate. But users can enter at any point.

Error Handling

For API-backed failures (quota, pricing, creator, monitor, campaign, collection, email, message, crm, brand-monitor, export), use the CLI response's action field when present:

• action.url — where the user should go
• action.hint — what to do

Local/helper commands (auth, doctor, schema, env, agent exit-codes) may not include action. Read their native output directly instead of assuming the API error envelope.

For unexpected failures, run doctor as a first diagnostic step.

References

• {baseDir}/references/cli-response-format.md — response envelope differences and error action handling
• {baseDir}/references/marketing-ops.md — campaign, collection, CRM, email/message, export workflows and mutation guardrails
• {baseDir}/references/brand-monitor.md — brand monitor routing, YouTube-only product signals, export boundaries
• {baseDir}/references/platform-support.md — data availability by platform
• {baseDir}/references/search-filters.md — filter selection by user intent
• {baseDir}/references/verdict-heuristics.md — detailed due-diligence rules and output structure