Player profile
STARK OS
An MIT-licensed, self-hosted team of seven role-based agents for social media work, each with its own prompt, tools, memory and preferred free OpenRouter model.
Profile
Where it fits
STARK OS runs locally as a FastAPI service with a Next.js dashboard. Seven agents cover a social media team's roles: a director (briefs, delegation, brand and security review), a copywriter, a community manager, a researcher, a video and visuals planner, a red-team agent and a publisher. Each agent is defined by a TOML file and a system prompt.
The director delegates work to specialists, and team reviews run the reviewers in parallel before the red-team agent challenges the combined result. Agent-to-agent messages are recorded and shown in the dashboard. A permission engine limits which tools each agent may call; the red-team agent is observe-only.
Drafts pass deterministic policy checks (platform length limits, leaked keys, personal data, missing ad disclosure, giveaway rules). Publishing uses human approval of the exact text or an explicitly enabled Autopilot trust level; edits cancel existing per-post approvals, and sensitive topics or failed checks still require review. Bluesky and Mastodon posting is real; other platforms produce drafts labelled as simulated.
Builder web analytics
Measure outcomes from STARK OS posts
STARK OS reports platform metrics for posts it actually published. Separately configured product analytics can show what those posts did for a linked site or product.
This is an optional workflow, not a native STARK OS integration with Agent Analytics. Configure tracking on the linked site and analytics access separately; this listing does not install tracking or report STARK OS agent runs to Agent Analytics.
First loop to measure
- the agent team drafts and checks a post, and a human approves and publishes it
- the post links to a separately instrumented site or product page
- a human or an agent with separately configured analytics access reads the observed results
- the team uses that evidence to plan the next campaign
Copyable prompt
If tracking and analytics access are already configured for the linked page, report measured visitor outcomes for the campaign window with their limitations. Otherwise, identify what is missing. Do not assume a native STARK OS integration or infer causation from a before-and-after comparison alone.