Menu
Route Goblin Recon requests to one primary workflow and run the smallest useful sequence.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Write platform-ready GenX captions and descriptions with brand guardrails. Use for caption, description, hook, LinkedIn post, Instagram caption, YouTube Shorts description.
Build Instagram and Facebook carousel assets from a reference-first workflow. Use when the user asks to make a carousel, generate carousel slides, create a social image, or build editable carousel layers.
Public competitor scan that turns sourced observations into cell-ready moves. Triggers on "run competitor scan", "check competitors", "competitor deep scan".
Generate and score outbound email subject lines, openers, and short drafts.
Goblin Recon router skill. Routes user commands to the smallest useful workflow: Social Pulse, Clip Mine, Clip Vault, Competitor Scout, Brand Gate, Email Hook, or Carousel Generator.
Extract 15-60 second clip moments from source video transcripts or user-provided excerpts.
| name | orchestrator |
| description | Route Goblin Recon requests to one primary workflow and run the smallest useful sequence. |
| category | genx-marketing |
| version | 1.1.0 |
Purpose: choose the workflow, set the stop condition, and avoid tool sprawl.
Core shape:
Router -> Workflow -> Tools -> Normalized Data -> Score/Gate -> Human Gate -> Memory
run fast scanrun deep social scanrun signal scanmanual scan this ...run social pulserun clip minefind clips about ...find the moment in ...what clips are readyrun competitor scanwrite email hooks ...run carousel generatorrun full scanrun full autonomous scan| Intent | Workflow | Stop when |
|---|---|---|
| Trends/hooks/formats/ideas | Social Pulse | ranked opportunities exist or sources fail |
| Source videos/timestamps | Clip Mine | 1-3 viable briefs or no safe source |
| Prior clips/status | Clip Vault | requested records shown/updated |
| Competitors | Competitor Scout | sourced report complete |
| Brand fit | Brand Gate | pass/revise/shelve returned |
| Email hooks/drafts | Email Hook | scored variants returned |
| Carousel/social image | Carousel Generator | local assets rendered |
If mixed, use the smallest chain. run full scan = Social Pulse, then Clip Mine for only the strongest 2-3 candidates.
run full autonomous scan = Social Pulse + limited Clip Mine + optional competitor/caption packaging from approved public/local sources. Still stop for secrets, paid setup, login/paywall/captcha bypass, publishing/sending, destructive deletes, deploys, or commits.
State selected workflow and scan mode before collecting data.
Before brand-facing output ask once:
Audience? B2C / B2B / Both
Destination? Faceless Instagram / personal brand / client work / internal / email / other
Tone? professional / casual / edgy / warm / wry / reflective / analytical / bold / platform-native
If skipped: default to Both / Faceless Instagram / professional.
scan mode -> collect sources -> social_intake -> scoring -> Security Rail -> report
Use skills/trend-radar/SKILL.md. Save final report to vault/reports/.
trend/topic -> source-hunter -> transcript -> clip_store dedup -> moment-finder -> brand_gate -> Security Rail -> Human Gate
Use:
skills/source-hunter/SKILL.mdskills/moment-finder/SKILL.mdgoblin_recon.tools.youtube_toolgoblin_recon.tools.clip_extractorgoblin_recon.tools.clip_storeClip duration: 15-60 seconds. No full raw transcript storage.
Query vault/clips.db first. Do not run trend/source collection.
Useful commands:
scripts/query_clips.py list --status approved
scripts/query_clips.py list --query "topic"
scripts/query_clips.py brief <clip_id>
Use skills/competitor-scout/SKILL.md. Every claim needs public URL + date. Run Security Rail.
Use skills/email-hook/SKILL.md and goblin_recon.tools.email_gate. Run Security Rail before delivery.
Use skills/carousel-generator/SKILL.md and goblin_recon.tools.carousel_renderer. Render locally; Replicate is optional fallback enhancement, not required.
Before user delivery, run skills/security-rail/SKILL.md on reports, briefs, claims, copy, and recommendations.
Decision handling:
APPROVE -> deliver
REVISE -> deliver revised safe version
SHELVE -> do not recommend
NEEDS HUMAN REVIEW -> label clearly
Every report/brief starts with ## Decision and includes:
No fabricated sources. When uncertain, shelve.