856 skills · sorted by relevance
Monitor machine learning training through WandB or training logs. Check for invalid numbers, worsening metrics, and stalled improvement, then report whether to continue, wait, or stop, preserving evidence and notes when stopping a run.
Understand a single arXiv paper through an AlphaXiv summary, with full text and LaTeX source available when needed. Use it to identify key findings or ask specific questions about methods, experiments, and equations.
Draft patent claims from invention disclosures, prior art reports, and novelty assessments. Create broad independent claims and narrower dependent claims, check their support in the specification, and adapt separate claim sets for China, the United States, or Europe.
Style materials to match the OpenAI and Codex look using specified colors and fonts. Apply distinct formatting to headings, body text, code, and shapes, with fallback fonts when the preferred ones are unavailable.
Improve MNN's Vulkan backend by separating CPU scheduling bottlenecks from GPU computation costs. It guides operator and shader changes, new feature integration with compatibility fallbacks, and device testing to verify both correctness and end-to-end performance.
Check whether the official Obsidian command-line interface is ready, then use it to read and search a selected vault. Access supported backlinks and tags, with direct file reading as a fallback when the interface is unavailable.
Creates original-language, translated, and bilingual SRT subtitle files from YouTube links, local videos, or other supported media. Uses platform captions when available, falls back to Whisper transcription, and provides subtitle output for short vertical videos.
Review supplier agreements against your team's contract guidelines. Produce a memo explaining deviations, legal risks, business impact, specific wording changes, fallback positions, and required approvers, so a lawyer can decide how to proceed.
Review AI vendor contracts against your organization's policies. Identify gaps in data training rights, liability, model changes, and other terms, then prepare targeted contract edits and escalation options for issues outside agreed limits.
Adjust an existing commercial contracts profile through a guided conversation. Update risk preferences, approvers, contract positions, or review formats one change at a time, with an explanation of how each update affects later work.
Deploy and operate Vercel projects from the command line. Manage environment variables and domains, investigate build failures with logs and metrics, and handle platform settings while checking which project and team each operation targets.
Create a single-column HTML email for a product launch, with a centered brand wordmark, hero image block, headline, copy, action button, specifications, and footer. Table-based layout and inline colors support compatibility with email clients.
Convert PDF, PowerPoint, Excel, and Word documents into structured Markdown for search and information extraction. Outputs include document metadata, a parsing confidence score, and the parser used, with fallback parsers available when earlier attempts fail.
Plan and verify OpenRig startup, shutdown, and recovery. Compare reported status with actual running sessions, distinguish recovery outcomes, and define reboot tests that expose partial failures instead of hiding them behind a fresh launch.
Review fiction for problems in structure, characters, prose, and consistency. Receive prioritized suggestions supported by passages from the text, using multiple reviewers when available or a single reviewer when the required setup is unavailable.
Add thumbnails, colored badges, and data-driven styling to TypeScript lists, cards, and panels. Coordinate component logic with CSS, include fallbacks for missing data, and check how the updated display handles empty lists and long text.
Recover failed writes of large files or code containing special characters by placing content in a temporary Python script. Use raw strings or Base64 encoding, then check that the output file exists and has content.
Save large text files when write_file or shell_agent encounters payload size limits. Create the parent directory first, write the content with Python through run_shell, and verify the resulting file.
Validate spreadsheet sources before running Python scripts to transform data, update multiple sheets, or apply formatting. Check required data first, try backup sources when needed, and report errors clearly if processing cannot proceed.
Recover from failed automated file creation with an explicit shell workflow. The skill guides agents through checking folders, writing and running scripts, and inspecting output files to confirm that deliverables were actually created.
Create documents when several research or execution tools fail. Delegate PDFs and complex layouts to shell_agent, a separate agent, or save Markdown directly, while distinguishing verified facts from general guidance and checking the output.
Create a structured Markdown report when multiple retrieval tools fail. Using available domain knowledge, the workflow separates verified facts from general guidance, marks missing data, and saves a document with limitations and practical next steps.
Recover failed Python tasks by saving the code as a script and running it through the system shell. The workflow covers spreadsheet generation, data processing, output verification, and troubleshooting missing packages, paths, or permissions.
Continue suitable tasks when web search repeatedly fails by assessing whether existing knowledge is sufficient. Mark unverified claims, distinguish confidence levels, and provide a follow-up checklist so readers can see what still needs checking.