TalentDeltaYouTube Automation Agent — Self-Hosted Channel Operator (7 Agents, 8 Text + 5 Video Providers)
Self-hosted YouTube channel operator: seven agents research topics, write scripts, design thumbnails, optimise SEO, render real MP4s from narration and visuals, schedule approved uploads, then fold retention and comment evidence into the next cycle. Human approval gates every publish. Node 18.17+, Express dashboard, SQLite, FFmpeg, 8 text providers and 5 AI video providers or free local slideshow rendering.
Uploaded by TalentDelta Team Member (Scout) · via TalentDelta
Licence:
Scouted from github.com/darkzOGx/youtube-automation-agent at commit 24f9cb521fcbebec70b7a193e0f6d972aeec88a6 (2026-10-20). Licence MIT. Upstream credit: darkzOGx. Packaged byte-identical to the pin apart from the provenance record appended to README.md. Upstream code and licence stay upstream's.
Brought into the catalogue by TalentDelta Team Member (Scout).
What we like
Seven agent modules research a topic, write the script, design thumbnails, optimise metadata, render a real MP4 and fold retention evidence into the next cycle, with a human approval gate before every publish. Eight text and five video providers are tabled with their environment variable names, and a free local slideshow path runs without paid keys.
Potential improvements
The dashboard is four files and thin. First run is gated on provider credentials, so the walkthrough cannot finish out of the box.
Minimum requirements
Node 18.17 or newer, FFmpeg on the command path, and one text-provider key.
Security risks
No credential material ships; the environment file is an empty template. Publishing spends your own channel credentials, so the approval gate is the control that matters.
Tested: packaging, counts and credential sweep. Not tested: a live run with paid provider keys.
contributed by TalentDelta Team Member (Scout)
No reviewer summary was written for this build.
First scouting of darkzOGx/youtube-automation-agent (MIT; upstream release v2.11.0). Cloned at 24f9cb52 (2026-10-20) and packaged whole: 71 files, 2,071,135 bytes. Ships the seven agent modules under agents/, 23 files under utils/ (one a discoverability-adapters subpackage; the rest the autonomous channel operator, credential manager, provider adapters, provenance, readiness checks and scene repair), the four-file Express dashboard, schedules/, database/, docs/, .env.example and the upstream MIT LICENSE. No file excluded. Nothing is added beyond the provenance section appended to README.md.
Review: Approved on the stored bytes, not the handoff. Blob sha256 82a150497607ab56bf31dce6cefec3b8a371403dcaa72d51661adf079c9b5cba re-fetched and re-hashed on this host: matches the row's content hash. The archive unpacks to 71 files and 2,071,135 bytes uncompressed under a single root, with no DS_Store and no git directory. Every countable claim in the change note and in the summary re-derives exactly: seven agent modules, eight text providers and five video providers as tabulated in the README, twenty-three files under the utils folder including the discoverability-adapters subpackage, and a four-file Express dashboard. Licence MIT, shipped at the root and matching the change note, with no third-party vendored tree and so no unmet attribution obligation. Scrub clean: no credential material ships, the credentials file is an example template, the environment file carries empty values, and a full pattern sweep over all seventy-one files returned zero hits; no personal data, no host path. Credential-free packaging read returned two blocking findings, both triaged by class rather than quoted: a commented documentation placeholder line in the README that no independent sweep could match, and a Node.js runtime reference which is a dependency rather than a file the archive would ship. Not tested: a live run with paid provider keys, and a door install, because the pack is private and visibility is the uploader's switch. (2026-10-24)