Skip to content

v1.2.0

Latest

Choose a tag to compare

@github-actions github-actions released this 08 Aug 17:26
· 416 commits to main since this release

v1.2.0 — 2026-08-08

Skills reach every agent, not just Claude Code

  • The ten skills moved from adapters/claude-skill/skills/ to a top-level
    skills/ directory, which is where the open skills ecosystem looks. They
    now install into any of its 27+ supported agents:
    npx skills add oxbshw/watch-skill -g. Buried where they were, only
    Claude Code could see them — the agent-facing layer that decides when to
    watch video was the least distributed part of the project.
  • The Claude Code plugin is rooted at the repository instead of a
    subdirectory, so one copy of each skill serves both ecosystems rather than
    a mirror that can drift. skills.sh.json declares the package.

Sixteen vision providers, from six

  • Added Groq, Together AI, Fireworks, DeepSeek, xAI, Mistral, Moonshot,
    Z.ai, and Qwen (DashScope) — plus custom for any OpenAI-compatible
    server: vLLM, LM Studio, llama.cpp, LiteLLM, Azure OpenAI, a company
    gateway. Two contributors asked for the last one before it existed.
  • These are registry entries, not code. Every one of them speaks OpenAI's
    /chat/completions, so a single builder is generated from the table and
    adding a vendor touches no request logic.
  • --base-url on setup-vision, and a WATCHSKILL_<PROVIDER>_BASE_URL for
    each, so one entry reaches a regional endpoint, a proxy, or a self-hosted
    server. Passing it to a provider with its own wire format is a clear error
    rather than a silently ignored flag.
  • WATCHSKILL_GROQ_API_KEY existed with no Groq provider behind it. The
    setting was read, reported by doctor, and impossible to use.

Three more agents

  • Zed, whose key is context_servers rather than
    mcpServers — a config copied from any other client here does not work.
  • Roo Code, with the project-scoped .roo/mcp.json
    a team can commit.
  • Continue, which reads one file per server from
    .continue/mcpServers/.

Supply chain and platform

  • PyPI uploads carry PEP 740 attestations. Publishing moved from
    uv publish to the official PyPI action, which signs the distributions
    with this workflow's identity — an installer can verify a wheel was built
    here and not swapped in transit.
  • The container image is multi-arch. linux/arm64 alongside amd64, so
    Apple Silicon and ARM servers stop running ffmpeg under emulation. The
    pushed digest also gets a signed build-provenance attestation, and the
    build emits an SBOM.
  • Dependabot watches Actions, Python dependencies, and the base image
    weekly. A compromised action is the shortest path into a pipeline that can
    publish to PyPI.
  • AGENTS.md at the repository root — the cross-agent convention that
    Codex, Cursor, Copilot, and others read. For a project whose users are
    agents, not having one was an odd gap.

Changed

  • An agent page no longer requires hand-drawn art to exist. The test
    demanded an avatar for every page, so adding an agent was blocked on
    someone illustrating it — which contradicted the twenty-minute
    contribution path in CONTRIBUTING. A page without art belongs in the
    matrix; it joins the README gallery when the art lands. Art that does
    exist must still be shown, and no page may be orphaned from the matrix.