AlphaGBM/skills

This page introduces an external open-source repository. It is not an HDATF product.

What it does

AlphaGBM Skills packages financial-research instructions and selected Python runners for use inside compatible AI workspaces. Users request stock research, options comparisons, news interpretation, report breakdowns, or comparison of supplied research snapshots. These packages are not one trading application, and method-reference packages do not expose a live data API. Callable packages include a Python 3.9 or newer runner, so skill installation and runtime preparation are distinct from connecting an AlphaGBM account. Published research reads can be public, while account-backed stock and options analysis requires a personal API key, permission to use the allowance, and applicable subscription access. Installing a package does not grant extra quota or access to private reports, and some workflow interfaces require a staged backend contract. Investment Review compares explicitly supplied, compatible versioned JSON snapshots locally without a key, network request, upload, or account-history access. Its output separates comparable changes, missing evidence, and judgments needing review; differing identities, versions, dates, or units can prevent meaningful comparison. Demo fixtures are not captured paid responses, research scores are not return guarantees, and these workflows do not place trades.

How it helps ATF

Its direct financial domain is not a core manufacturing function, but the evidence-comparison design is a candidate reference for LabChin research revisions and Company Brain report updates. A proposed comparison record would retain baseline and current identifiers, schema version, observation date, source fields, and an explicit reason when values are not comparable. The financial runner should not be applied unchanged to company documents; those records would need their own schemas and comparison rules. Harness could separately record public reads, approved chargeable requests, and recoverable task identifiers rather than retrying an uncertain paid call. Evaluation should use authorized sample snapshots with changed units, stale dates, missing fields, and mismatched subjects to confirm that unsupported differences stay unavailable. Financial forecasting, automatic trading, and treating synthetic demos as real evidence are outside this proposal, and no product connection or research accuracy has been verified.

License

MIT Permissive. Commercial use and changes are allowed if the copyright notice is kept.

More in this category

• earthtojake/text-to-cad
  text-to-cad gives existing coding agents skills and a local runtime for creating mechanical CAD models. Users provide descriptions, dimensions, pictures, drawings or existing CAD files, and the agent writes editable Python design code. The code uses build123d and an Open CASCADE-based geometry kernel rather than a bundled text-to-3D model. Outputs include STEP for CAD exchange and GLB, STL or 3MF for viewing and downstream fabrication workflows. The plugin also supplies a viewer, engineering drawing workflows and guided design-for-manufacturing checks. Starting requires a compatible agent, uv and downloaded Python dependencies, with plugin and viewer requirements depending on the chosen client. Generation quality depends on the agent, and the manufacturing checks are guidance rather than certification. Re-exporting STEP can lose manufacturing annotations and geometric tolerances, while references to selected faces are valid only for the saved model version from which they were obtained.
• jwangkun/claude-for-financial-services-cn
  A set of Claude skills adapted for China's A-share financial market. It covers company valuation, financial statement models and research reports using local data sources, accounting standards and report formats.
• chengyi-ai/native-subtitle-quote-image
  This Agent Skill and its Python tools help content editors turn real video frames into quote images without confusing original subtitles with later additions. Inputs are an authorized local video or YouTube URL, selected timestamps, and optionally reviewed lines in JSON. Native mode crops and joins subtitles already burned into the video pixels rather than redrawing recognized text. Script mode draws approved lines on real frames and must be identified as added subtitles. Outputs include JPG images, the corresponding timestamp or line JSON, and a contact sheet for multi-image work. The workflow samples candidate frames, checks subtitle boundaries, composes the image, and requires visual review of each result. Getting started requires installing the skill in a compatible agent and separately preparing Python 3.10 or newer, Pillow, and imageio-ffmpeg or FFmpeg; the scripts can also run without an agent. YouTube acquisition additionally needs yt-dlp and a supported JavaScript runtime, while Chinese, Japanese, or Korean scripted text needs suitable fonts. Switchable subtitle tracks cannot become native subtitle pixels, and neither enlargement nor the MIT code license supplies better source detail or rights to third-party footage.
• danyuchn/asd-ste100-skill
  This Claude Code skill applies ASD-STE100 Simplified Technical English rules to ambiguous agent-facing English. It focuses on short sentences, consistent word meanings and one instruction per sentence.
• cclank/lanshu-create-ai-presenter-video
  This Agent Skill guides a file-reading and shell-capable agent through explainer-video production rather than providing a standalone presenter model. A presenter job takes a topic or script and an authorized adult portrait, while the styled route can work without a presenter image. The presenter route coordinates narration, presenter generation, lip synchronization, captions, editing, and review; the styled route uses reusable visual styles and timed narration. Approved narration supplies the timeline, and a job JSON records approvals, capabilities, task identifiers, artifacts, and review reports. Bundled scripts derive production states from recorded evidence and prepare a master video, share copy, and delivery report with media checks. Setup requires a compatible agent, Python 3.9 or newer, FFmpeg and ffprobe, plus the documented shell tools; installing the skill does not supply generation services. Presenter production needs separately available voice, presenter, and lip-sync capabilities, while styled production adds Node.js and HyperFrames and uses MiniMax voice access or the user's own narration. Paid calls and uploads need approval, and task identifiers and retry limits are used to control repeated charges. Technical checks do not prove factual accuracy or visual quality by themselves, and human review remains part of the workflow.

Only repositories in the ranked Trendshift lists are included, and the lists are used only to find candidates. We do not copy their ranks. Descriptions, licenses and star counts come from each GitHub repository. The notes are our own reading. We have not tested these projects, and a place on a trending list does not prove quality.