Kelly creators

Kelly Creators

Overview

Use this skill as Kelly's personal influencer/creator-marketing operator. It keeps a file-backed App-in-Skill dashboard over creator candidates, outreach drafts, campaign briefs, and ROI, plus a review queue of agent-drafted outreach and pre-publication content gates. The skill sweeps candidates from whatever Kelly feeds it — a niche, a brand brief, a competitor's creators, an exported list — scores fit, drafts outreach and briefs, and executes approved sends only through other channels (for example instagram-outreach or kelly-email) after explicit approval.

An item is a creator engagement: handle, platform, followers, engagement_rate, fit_score, niche, est_rate, proposed_action, suggested_reply, and the standard status/decision/execution blocks. A second item type is a quality gate on a live creator's draft post.

Default interaction mode: App UI. Unless the user explicitly asks for chat-only handling, check onboarding/config, refresh or regenerate the local creator snapshot, start/reuse the local app with app/start.sh, and give the actual local URL. Use chat-only mode only when the user says "纯聊天", "chat only", "不要打开 UI", or similar; in that mode present numbered creators (#1) and take verdicts in the conversation.

Philosophy

This skill is an App-in-Skill: a Codex/agent skill paired with a small local companion UI. The skill does the real work (external reads, reasoning, drafting, executing approved sends); the app is a quiet local operator surface that reads and writes local files only and never performs an external side effect. See the App-in-Skill specification paper: https://mr-kelly.github.io/research/app-in-skill-specification-for-pairing-agent-skills-with-a-local-companion-ui.pdf.

The domain follows the four-phase influencer-marketing discipline — Discover → Plan → Activate → Measure — used both as the pipeline funnel and as a phase facet on every item. Human clicks are reserved for judgment, edits, exceptions, and irreversible or money/contract actions.

Capabilities (Discover / Plan / Activate / Measure)

The agent covers sixteen sub-skills, grouped by phase. Each generated item is tagged with its phase so the desk always shows where the program stands.

• Discover — audience-mapper (who the brand's buyers are), trend-spotter (rising formats/sounds/topics), influencer-discovery (sweep candidate creators), fit-scorer (the objective C³ ACE match score).
• Plan — competitor-tracker (which creators competitors use), campaign-planner (goals, mix, timeline), brief-generator (creative brief per creator), budget-optimizer (allocate spend across the roster).
• Activate — outreach-manager (draft and track outreach DMs/emails), content-reviewer (the pre-publication SHIP/FIX/BLOCK quality gate), contract-helper (terms, usage rights, exclusivity), content-amplifier (whitelisting/boosting approved posts).
• Measure — landing-optimizer (post-click experience), performance-analyzer (reach/engagement/conversion), roi-calculator (spend → estimated value, CPM, ROI), report-generator (campaign wrap-ups).

Fit score — C³ ACE

fit_score (0-100) is the objective matching score: Content × Community × Credibility crossed with Audience × Cost × Engagement. fit_breakdown carries the six sub-scores shown in creator detail. Prefer fit over raw follower count when ranking candidates.

Content-reviewer quality gate

Before a live creator's post publishes, the agent runs a content-reviewer gate that outputs SHIP / FIX / BLOCK over ftc_disclosure (is #ad disclosed above the fold?) and claim_authenticity (are product claims supportable, no cure/medical overreach?), plus brand_safety. A fix or block verdict is surfaced in the outreach queue for a human decision. This gate is item_type: "quality_gate" with gate_verdict and gate_checks.

App UI Screenshots

<table> <tr> <td width="50%"><img src="assets/screenshots/overview.webp" alt="Kelly Creators overview"></td> <td width="50%"><img src="assets/screenshots/creators.webp" alt="Kelly Creators candidates"></td> </tr> <tr> <td><strong>Overview</strong><br>Pipeline funnel across the four phases, budget allocation, total reach, and the top fit-scored candidates.</td> <td><strong>Creators</strong><br>Sortable candidate cards with C³ ACE fit scores, platform, niche, and audience size.</td> </tr> <tr> <td width="50%"><img src="assets/screenshots/outreach.webp" alt="Kelly Creators outreach queue"></td> <td width="50%"><img src="assets/screenshots/roi.webp" alt="Kelly Creators ROI board"></td> </tr> <tr> <td><strong>Outreach</strong><br>Needs-review approval queue with editable outreach drafts and the FTC/claim disclosure gate.</td> <td><strong>ROI</strong><br>Per-creator spend, estimated value, CPM, and return once a partnership goes live.</td> </tr> </table>

Boundary

• The skill may read sources Kelly provides, sweep and score creators, draft outreach/briefs/contracts, run content gates, validate schemas, and write local handoff files.
• The app reads and writes local files only. It must never send DMs or emails, call platform APIs, post content, or perform any external side effect.
• Outbound outreach, briefs, and contracts are always approval-required. Sending is delegated to other skills (for example instagram-outreach, tiktok-outreach, kelly-email) and happens only after the user approves the specific item. scripts/execute_decisions.ts only records handoff operations in execution_report.json; it performs no sending itself.
• Treat money and contract terms (rates, usage rights, exclusivity) as sensitive and approval-required. Do not commit config.local.json, env files, app/.data/, exports, or creator contact details.

First Run And Onboarding

On invocation, check app/.data/onboarding.json and private config readiness. If onboarding is absent/incomplete, guide setup before doing real creator work.

Private config priority:

1. KELLY_CREATORS_CONFIG=/absolute/path/to/config.json
2. skills/kelly-creators/config.local.json
3. ~/.config/kelly-creators/config.json
4. skills/kelly-creators/config.example.json as template only

Env priority:

1. Existing environment variables
2. KELLY_CREATORS_ENV_FILE=/absolute/path/to/.env
3. Repository root .env
4. skills/kelly-creators/.env.local
5. ~/.config/kelly-creators/.env

Ask for non-secret setup details only: operator profile (name, role, company, timezone), brand(s) and positioning, target niches, program budget and base currency, outreach platforms and which skill handles each, style/tone for drafts, risk keywords (money/contract), and which env var names hold platform tokens. Never ask the user to paste secret values into chat.

When setup is complete and the user confirms, write app/.data/onboarding.json:

{ "completed": true, "completed_at": "ISO timestamp", "config_version": "1" }

Local App

Start the dashboard with:

skills/kelly-creators/app/start.sh

The app uses local HTTP on 127.0.0.1, preferring port 3200 through 3999, or KELLY_CREATORS_UI_PORT when set. The launcher reuses a running instance only when /api/state proves it is the same app (app: "kelly-creators").

Required app views:

• #/overview: command desk. Human-attention counts, the discovery → outreach → negotiating → live → measured funnel (tagged by phase), budget allocation, total reach, and top candidates by fit.
• #/creators and #/creators/<creator_id>: candidate cards with fit score, platform, niche, followers, engagement rate, and rate — sortable by fit / followers / engagement / cost. Detail shows the C³ ACE breakdown, the outreach draft (or the quality-gate checks), and the full engagement record.
• #/outreach: review queue over agent-drafted outreach and content gates in workflow states needs_review, changes_requested, approved, done, blocked. Each item shows a stable row ref (#1), fit score, phase/risk badges, an editable suggested_reply draft, a Review note textarea, and decision buttons Approve / Request changes / Block that write to decisions.json. The queue is read-only while agent.lock exists.
• #/roi: per-creator spend, estimated value, CPM, and ROI.
• #/settings: sanitized config summary. Operator profile, brand(s), program budget and target niches, configured platforms, env readiness booleans, data provider name, and onboarding state. Never expose secret values.

Demo mode:

• ?demo=1 opens a deterministic mock program for documentation and screenshots.
• ?demo=overview, ?demo=creators, ?demo=outreach, ?demo=roi, and ?demo=detail select named mock scenes; detail deep-links to a creator detail.
• lang=en or lang=zh forces UI chrome language for screenshots.
• Demo API responses must never read or write files under app/.data/ or any private config. All handles are invented.

UI language: support English and Chinese chrome with Auto default. Keep creator names, handles, notes, and drafts in their original language.

File Contract

Read references/creators-schema.md before editing the app, scripts, or any generated creator JSON.

Primary local files:

• app/.data/creator_snapshot.json: normalized creator snapshot (creators, metrics, funnel, warnings) generated by the skill/scripts.
• app/.data/decisions.json: user verdicts and review notes keyed by creator_id, written by the app.
• app/.data/agent_tasks.json: queued agent work — engagements in changes_requested with the user's comment. The skill polls this to pick up revisions.
• app/.data/execution_report.json: latest handoff/execution results written by scripts/execute_decisions.ts.
• app/.data/onboarding.json: onboarding completion marker.
• app/.data/agent.lock: temporary lock while the skill is generating or executing. The app rejects decision writes while it exists.
• config.local.json: private operator configuration, ignored by git.

Use scripts/validate_ui_schema.ts before relying on a snapshot in the UI. The app may show an empty setup state when no snapshot exists.

Normal Workflow

1. Detect mode. Default to App UI.
2. Load private config through the config helpers. If only config.example.json exists, enter onboarding.
3. Discover/Plan: when Kelly feeds a niche, brand brief, competitor, or candidate list: acquire app/.data/agent.lock, update creator_snapshot.json — upsert creators by stable ids, compute the C³ ACE fit_score and fit_breakdown, set stage/phase, draft suggested_reply outreach or briefs with status: "needs_review", add money/contract risk badges and est_rate, recompute metrics — validate with scripts/validate_ui_schema.ts, then release the lock.
4. Start/reuse the UI and report the URL so Kelly can review the funnel, the candidate cards, and the outreach queue.
5. Poll app/.data/agent_tasks.json for changes_requested items. Re-draft each per the user's comment, set it back to needs_review, and clear the task.
6. Activate: on "execute" / "send approved outreach": re-read decisions.json, re-check the lock, and run scripts/execute_decisions.ts --apply to record send_outreach / send_brief / draft_contract operations in execution_report.json. Then perform the actual sends only through the corresponding skill with the approved, possibly user-edited draft, one item at a time, and mark each done afterwards. Run the content-reviewer gate on live drafts and surface FIX/BLOCK verdicts for approval.
7. Measure: as campaigns go live and complete, update spend, est_value, and cpm and move engagements to measured; summarize ROI in the #/roi view.
8. Never send anything for items without an explicit approve decision, and never re-send items already marked done in the execution report.

Safety Defaults

• Treat every outbound message, brief, contract, rate commitment, usage-rights term, and exclusivity clause as approval-required.
• Never publish or approve a live post that fails the content-reviewer gate (missing FTC disclosure or unsupportable claims) without a human decision.
• Store only the minimum creator content needed for review; keep raw scraped data and platform exports out of the snapshot.
• Redact tokens and credential-like strings from logs, reports, and UI state; expose only boolean readiness for configured env vars.
• Keep stable ids (creator_id) and ref numbers so repeated updates and executions are idempotent.
• If decisions and the snapshot disagree (missing creator, stale ref), stop and regenerate rather than guessing.