Skill packs › Skill

Content AgentIN EVERY PACK

The Content Agent – drop in one raw video and get back a lightly edited YouTube upload (private, for your review), a blog draft with schema, 5-10 short-clip picks, platform social posts, and an email – everything grounded in YOUR transcript and YOUR files, nothing invented, nothing published without your click.

Skill file content-agent.md · last updated Aug 2, 2026

Use this when you have ONE raw recording — a talking-head video, a podcast episode, a webinar, a livestream, even a long voice note — and you want it to become a week of distribution without you touching an editing timeline. Run it weekly. This is the Content Factory’s big sister: Content Factory writes the words; the Content Agent also handles the video itself and stages everything as drafts.

Read these first, every run: your brand-voice document, your ideal-client file, and your links file (your real domains, offer pages, and social URLs). Everything below is grounded in those plus your transcript — and nothing else.

The promise

One recording in. Six things out, all as drafts for your review:

1. A lightly edited video, uploaded to YouTube as Private (or staged as an upload kit).
2. A blog post draft (2,000–3,500 words) with the video embedded, a 3-line TL;DR, and the cleaned transcript at the bottom.
3. 5–10 short-clip picks with timestamps, hooks, and per-platform captions.
4. 3–4 platform-native social posts (LinkedIn · Instagram carousel script · X thread · Threads).
5. One email to your list, in your newsletter’s voice.
6. Schema + internal links: JSON-LD for the post, plus 3–5 proposed internal links from your existing pages.

Hard rules (the whole reason you can trust it)

• Draft-only. YouTube = Private. WordPress/GHL = Draft. Email = saved file, never sent. The agent does not publish; YOU publish. Verify the status after every upload and say what you verified.
• Nothing invented. Every claim, quote, name, number, and link comes from your transcript, your knowledge-base files, or a page actually fetched and confirmed live. If a link isn’t in your files and doesn’t resolve when checked, leave it out and say so.
• Quotes are verbatim from the transcript. Image/thumbnail captions describe only what is literally in frame.
• Your voice, not AI voice. Use the brand-voice file. Style gate: no “not just X, but Y” constructions, H2s start with verbs where natural, no sentences ending in prepositions, no em-dash soup, TL;DR liftable word-for-word. Run the gate as a mechanical final pass AFTER all edits and expansions — violations sneak in precisely when you lengthen or rework a draft.
• Re-ground every file against transcript.md, never against earlier drafts — one imprecise word in an early file otherwise spreads to all of them. Before handing over, spot-check three claims and every quote directly against the transcript.
• Log every run. Append one line per video to Outputs/processing-log.md: date · source file · what shipped · what’s waiting on you. Never process the same video twice — check this log first.

Folder convention

Work from three folders — Knowledge Base (voice, ideal client, links), Raw (new recordings go here), Outputs (everything produced). Google Drive or local, same names. New video = any file in Raw not yet in the processing log.

Inputs

• One raw recording from Raw: a video (best), audio, a YouTube link to something already uploaded — or even just a transcript or long text (the written five still ship).
• Your knowledge-base files: brand-voice, ideal-client, links file.
• Optional: a Descript account (Free covers the first full run) or your own editor — the agent never blocks on tooling.
• Optional: WordPress/GHL access for draft staging — paste-ready HTML works without it.

Steps

1. Intake. List Raw, diff against the processing log, pick the newest unprocessed recording. Confirm: “Processing [filename] — right one?” Load the knowledge base. Keep outputs in the video’s language unless told otherwise.
2. Light edit — pick the path you have, never block on tooling:

• Descript (account / connector available): create a project from the raw file, then run exactly: “Apply Studio Sound to the whole composition, remove filler words conservatively so it still sounds human, remove silences over 2 seconds, and render/export the result.” Nothing fancier. Then report receipts: fillers removed, trims made, before/after duration. Zero fillers found = zero removed is a valid result — never invent work. (Measured reality from our own runs: a 1-minute video edits in ~4 minutes for ~9 AI credits; a full HOUR of raw video edited in ~13 minutes for 49 credits — 35 fillers and 15m39s of dead air removed, 59:48→42:21. Descript’s free plan covers a first full run but watermarks video exports — fine for testing, not publishing.)
• Your own tool (CapCut, Final Cut, etc.): use the 3-line edit checklist (filler pass · silence trim · loudness normalize) and continue with the raw file meanwhile.
• No tool: proceed with the raw video unedited and say so — a real upload this week beats a perfect one next month.

1. Transcript. Export from Descript; or pull auto-captions if already on YouTube; or work from a transcript provided. Save Outputs/<video-slug>/transcript.md (timestamped where available).
2. Target keyword. From the transcript’s strongest theme + your positioning: one keyword you can actually win (specific beats glamorous). One line of why.
3. Blog draft. Restructure the transcript into an article in your voice: hook open, H2/H3s, verbatim quote callouts, 3-line TL;DR up top, key takeaways, video embed placeholder, cleaned transcript at the bottom. Wire in your offer/lead-magnet link FROM YOUR LINKS FILE. Passes in this order: outline → draft → expand → tighten → style gate LAST.
4. YouTube upload package. SEO title (≤70 chars, keyword-front), description opening with the meta title + 2-line meta description, then summary + timestamped chapters + verified links, then transcript excerpt. Tags. Thumbnail brief: 3 concepts, each = frame-grab suggestion + ≤4 overlay words. Upload as Private — via YouTube Studio in your signed-in browser, or the one-screen upload kit (youtube-upload.md) if you’d rather click yourself. Never public.
5. Clips plan. Scan for 5–10 self-contained 30–60s moments; score each: hook-in-1.5s · stands alone · quotable. For each: timestamps, the hook line, and captions for Shorts / Reels / TikTok / LinkedIn in that platform’s tone.
6. Social + email. LinkedIn post (200–300 words), Instagram carousel script (5–7 slides), X thread (6–10), Threads one-liner — each pointing to the blog post. Email (200–400 words) in your newsletter format: headline takeaway, one verbatim quote, one link.
7. Schema + internal links. JSON-LD (BlogPosting + Person + VideoObject) for the post. Crawl your site for 3–5 real pages that should link to the new post; propose natural anchors; touch nothing until approved.
8. Stage drafts + hand over. WordPress/GHL draft if connected (verify “draft” in the response); otherwise paste-ready HTML. Save everything to Outputs/<video-slug>/. Update the log. Close with the review list: “3 things to approve: YouTube (private) → publish · blog draft → publish · email → send.”

Output

• Outputs/<video-slug>/ with: transcript.md · blog-post.md (+ .html) · youtube-upload.md · clips-plan.md · social-posts.md · email.md · schema.json · internal-links.md.
• The lightly edited video on YouTube as Private (or the upload kit), with receipts: what changed, before/after duration.
• The WordPress/GHL draft if connected — status verified as “draft” and said so.
• One appended line in Outputs/processing-log.md.

Definition of done

• You can review everything in under 30 minutes and publish with clicks, not edits.
• Zero invented facts, quotes, links, or offers — three claims spot-checked against the transcript before handover.
• YouTube is Private, blog is Draft — verified and stated.
• Every asset points home (blog ← video ← clips ← social ← email), in your voice, in your language.
• The log line exists. Next week’s run knows what this week did.

Notes

• Weekly rhythm: record once → drop it in Raw → run this agent → review over coffee → publish. 52 recordings a year becomes 52 posts, ~500 shorts, ~150 social posts, 52 emails.
• What the edit does NOT do (say it out loud, it prevents heartbreak): no b-roll, no music, no burned-in captions, no jump-cut style, no color grade, no auto-thumbnails. It gets a raw recording over the publish line; it is not a video editor with bells and whistles.
• Model note: plans on the big model, drafts on the fast one (see model-judgment). No special setup.

Pairs with

→ content-factory (words-only weeks) → video-repurposing-agent (this is record-side; that watches the channel publish-side — the full loop) → definitive-article-writer (when a video deserves the canonical page) → dollar-a-day-strategist (put $1/day behind the winner) → recursive-self-improvement-qa (grade the run, better next week)

Built by Dennis Yu (BlitzMetrics / Local Service Spotlight). Reads your brand-voice + ideal-client + links files so everything sounds like you and points home. Draft-only by design: the agent prepares, you publish.

Learned in the field

Appended automatically by the self-improvement loop (Skill-Learnings/): dated lessons from real runs. Newest at the bottom.

<!– learning:ghl-mcp-truth-2026-07-27 –> July 27, 2026

HighLevel’s official MCP (https://services.leadconnectorhq.com/mcp/, Private Integration Token auth) exposes 36 tools — contacts, conversations, opportunities, payments, calendars, forms, social posts, blog posts, email templates. It does NOT expose funnels or landing pages, and the underlying REST Funnels API is read-only (list funnels / list pages / count pages only). There is no create or update endpoint in any version.

Consequence for every agent that touches a CRM: never promise to “build the funnel page.” Write the page to the client’s own WordPress site as a draft and hand over paste-ready copy for their page template. This is also the only route that works for clients not on the coach’s platform.

Send safety: conversations_send-a-new-message is a REAL send — never call it on a scheduled run. emails_create-template is the safe way to stage a daily email.

<!– learning:2026-07-29-edit-the-generator-not-the-page –> July 29, 2026 (from: applying the SEO-tree block across 13 surfaces on 6 domains, July 29, 2026)

Before editing any live page, ask what regenerates it

Twelve pages needed the same block. Two of them — localservicespotlight.com/skill-packs/ and /asset-tracker/ — are rebuilt from scratch by their own Python generators on a schedule. Editing those live would have looked like a clean success: HTTP 200, block in the rendered DOM, verification green. The next scheduled rebuild would have erased it, silently, and the job that erased it would still have reported success.

So the work split three ways and each page went through exactly one owner: generated pages got the block added to their generator, plain pages through REST, base64 payload pages by decoding, inserting, re-encoding. One owner per file, always.

Rules:

1. Establish the owner before the edit. Search the project for the page’s slug. If a script

writes it, that script is the only thing allowed to write it.

1. Two writers on one artifact is a bug even when both succeed. The loser is whoever ran

first, and nobody finds out until a reader notices something missing weeks later.

1. When a live page and a local template have drifted, do NOT “restore” from the template.

Our trunk page had picked up three hand-added blocks since its last build; republishing the template would have deleted all three. Diff first, then make surgical marker-guarded edits.

1. Every surgical edit replaces exactly once and raises on zero matches. A sub_once() that

throws when its anchor is gone turns silent drift into a loud failure the same morning.

1. Mark generated blocks with a data attribute, not a comment. data-system-tree="<node>"

survives every filter WordPress runs over content, and gives the checker something exact to assert and the updater something exact to replace.

Learned July 29, 2026.

Other skills: ai-search-visibility · boil-the-ocean · client-access-checklist · client-relationship-cadence · content-factory · definitive-article-writer

The full run order is on the skill-pack directory. Every skill here is one task from the Task Library — the library is the catalogue of what can be done; a pack is the subset you install; an agent is who runs it.