Jev Social v0.1.4 — explicit media intent and browser-grounded research

[IRONICBo]

Jev Social v0.1.4 is live: Jev chooses bounded social-research operations, the local socai CLI executes them in your signed-in browser, and the UI streams post cards, video evidence, and a source-linked report.

Try the live overview · View the source · Read the v0.1.4 release

What changed in v0.1.4

• Generic TikTok research no longer receives a media-download action. Explicit download intent is recorded before --download-media can run.
• SECURITY.md now documents the exact model payload, key and cookie boundaries, local retention, cleanup, cost, cancellation, and recovery behavior.
• The versioned demo and Agent Skill paths are backed by an immutable release, 57 passing tests, zero npm audit findings, and a successful real Pages deployment.

Install the Agent Skill

GitHub CLI 2.101 or newer can install the versioned skill for Codex:

gh skill install socai-io/jev-social jev-social@v0.1.4 --agent codex --scope user

The repository also remains compatible with the cross-agent Skills CLI:

npx skills add socai-io/jev-social --skill jev-social

Run the local demo

Node 20+, an OpenRouter key with Jev access, and a compatible socai CLI are required.

npx github:socai-io/jev-social#v0.1.4 onboard
npx github:socai-io/jev-social#v0.1.4

What happens during a run

1. Jev chooses an Instagram, TikTok, or LinkedIn route.
2. After every result, the app rebuilds a typed action space from observed posts, profiles, and URLs.
3. Jev chooses one read-only operation; socai validates and executes it in the local browser.
4. Captured evidence stays visible while the research report streams in.
5. Login gates, failed decisions, and step limits return explicit partial results.

Jev never writes selectors, shell commands, coordinates, or arbitrary browser actions. The skill also suppresses raw run JSON, credentials, executable paths, and local artifact paths.

Recorded evidence

Run	Result	Elapsed time
Instagram Jev research loop	Four source-linked records after two searches and one detail read	63.969 s
TikTok CLI search	Five public result URLs	7.570 s
TikTok video detail	Downloaded MP4 and poster; comments unavailable	38.263 s

These are individual recorded runs, not a benchmark or guaranteed latency. The TikTok search timing is socai-only rather than a complete Jev loop.

Current reliability work is tracked in the pinned roadmap and the v0.2.0 — Reliable research loop milestone. Focused issues and pull requests are welcome.

If this browser-grounded research pattern is useful, star Jev Social and try it against a real question.