Skip to content

v0.1.0

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 04 Sep 06:39
· 14 commits to main since this release
v0.1.0
f3cb1c1
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

First public release.

Added

  • Browser-driven TikTok collection that reads the feed responses the page's own
    JavaScript makes, rather than reimplementing request signing. Real Chrome,
    logged out, human-paced.
  • Local SQLite library (node:sqlite) holding every post, account, run, and
    analysis. FTS5 search over everything collected.
  • Run scoping: one research question per run, so unrelated niches never mix in
    an answer.
  • MCP server (26 tools) and a matching CLI. The server never calls a model —
    read_slides hands slide images to whichever agent is driving.
  • Metrics built to avoid being fooled by raw views: views-per-follower, outlier
    score (best ÷ median), a repeatable gate that needs 3+ hits with one inside
    90 days, save-rate calibration, and operator-cluster detection on shared
    sounds.
  • swipekit and swipekit-plan skills carrying the evidence floor and
    how to read the numbers.
  • Self-contained HTML report and a Notion-importable markdown + images export,
    both named after the run they cover.
  • One Chrome across every session: a second process attaches to the first one's
    browser over CDP instead of fighting it for the profile lock, and opens a fresh
    one if that session has gone away. Servers exit with the client that started
    them, and concurrent writers wait on the database rather than failing.
  • An optional HTTP transport for several agents sharing one warm cache. It binds
    to localhost and validates Origin, which is what actually keeps a web page
    out of a local server.
  • Each skill is carried as an independent copy per agent: .claude/skills/ for
    Claude Code, .agents/skills/ for Codex (with an agents/openai.yaml
    declaring the MCP dependency).
  • Depends on playwright-core, not the full playwright package: the tool drives
    your real Chrome (channel: "chrome") and never needs Playwright's ~1.5 GB of
    bundled browsers, so first install is seconds rather than minutes.
  • CI runs typecheck, lint, and test as separate jobs on every push and PR. A
    v* tag runs the gate and publishes a GitHub Release with notes from this
    file.
  • SECURITY.md, CODEOWNERS, and GitHub issue and PR templates.

Known limitations

  • Analyzes slideshows only. Video is collected with full stats but not yet
    interpreted.
  • TikTok only.
  • The HTML report's slide thumbnails need macOS (sips); everything else is
    cross-platform.
Assets 2
Loading