Data analysis, machine learning, LLM apps, and prompt design (9.1K skills)
Use Google Gemini to analyze audio, images, videos, and PDFs, or generate and edit images. Get results such as timestamped transcripts, summaries, extracted tables, answers about media, and images made from prompts.
Create or improve Claude skills with guidance on SKILL.md structure, reusable scripts, reference files, and packaging. The skill helps turn a task into practical agent instructions and resources, then refine them after use.
Build a persistent wiki from documents, papers, notes, or project documentation, then retrieve reusable evidence with source citations for agent tasks and recurring questions. It requires the llmwiki CLI or a connected llmwiki MCP server.
Set up an app measurement plan, choose useful events and KPIs, and interpret performance data across acquisition, engagement, retention, and revenue. It helps identify tracking gaps, organize dashboards, and turn metric trends into specific product or marketing decisions.
Analyze an app’s actual App Store Connect data synced through Appeeky Connect, including downloads, revenue, in-app purchases, subscriptions, trials, and country breakdowns. Compare time periods, spot trends or anomalies, and identify follow-up actions based on the app’s performance.
Set up or troubleshoot app install attribution across iOS and Android. Review privacy-aware signals such as SKAdNetwork and AdAttributionKit, MMP integrations, Google Play Install Referrer, conversion values, and deep links to clarify how campaigns connect to installs and revenue.
Analyze tau2-bench task exports to pinpoint where an agent diverged from expected actions. Trace tool calls by conversation round, explain likely causes, check whether relevant knowledge-base documents were available, and cite transcript and source evidence.
Run and troubleshoot model inference on Fireworks AI through its direct API or the Leanmcp gateway. Compare models, configure SDK or HTTP clients, handle reasoning and tool-call responses, and test latency, concurrency, and failures across model runs.
Adds file-based tracking to LLM and agent experiments, recording conversations, tool calls, results, traces, and token or cost data. Includes viewers and comparison reports to inspect failed runs, replay episodes, and compare models or experiment variants across supported platforms.
Use Tinker to fine-tune models with supervised or reinforcement learning, then sample from, resume, or export checkpoints. The skill also covers inference, token-level training data, hyperparameters, and estimating run costs from Tinker's pricing information.
Diagnose why a prompt produces weak or inconsistent results, then rewrite its task, context, and output format. Create reusable prompts, compare different prompt styles, and retest when moving between AI models.
Plan and assess data migrations by profiling real source data, documenting field mappings, rehearsing at full volume, and reconciling counts and values. Use it to choose a cutover approach, manage exceptions, investigate incorrect results, and create an executable rollback plan.
Create or revise agent skills with clear triggers and practical instructions. Review descriptions, diagnose skills that load too often or too rarely, check for overlap in a library, and test whether example requests select the skill appropriately.
Use this skill to clarify what metrics mean, assign dataset owners, set access expectations, and check data quality and lineage. It helps teams resolve conflicting numbers, identify authoritative datasets, and make shared data easier to use responsibly.
Use this skill to define an AI system’s intended use, evaluate its errors across user groups, plan monitoring and meaningful human review, and document its limits. It supports deployment and retirement decisions, especially when AI could affect consequential outcomes for people.
Design and troubleshoot data pipelines that ingest, transform, and deliver reliable data. Get guidance on batch or streaming choices, safe reruns, delayed or duplicate records, data quality checks, and investigating unexpected changes in dashboard numbers.
Build reports and dashboards around decisions people need to make. Structure metrics as drivers and outcomes, provide trusted self-serve analytics, show useful comparisons and data freshness, and review reporting so unused dashboards do not multiply.
Set direction for how an organization defines, governs, stores, and shares data. The skill helps resolve conflicting metrics, assess data quality for decisions, shape analytics capabilities, and establish governance for AI models, including what may be deployed and how it is monitored.
Design warehouse layers, table grain, dimensions, and shared metric definitions for analytics. Use it to model new data sources, restructure marts, or investigate inflated results and slow queries, with guidance on history and query performance.
Answer business questions with careful data analysis: define the decision, choose a meaningful comparison, inspect data quality, and check for common statistical traps. Report estimates, uncertainty, and limits so teams can review evidence before acting.
Inspect large data files before reading their contents into an agent’s context. The skill returns a compact pointer with schema, shape, samples, and key values using bounded memory, so the agent can select specific rows, columns, or ranges to read.
Run compute jobs that need cloud CPUs, GPUs, or elastic capacity on Modal. The skill writes a function in the workspace, runs it with the user's authenticated Modal CLI, and saves outputs and run details for later traceability.
Use this skill to run social science analyses or read Stata and SPSS data. It reports estimates and uncertainty, checks code against a preregistration plan, verifies fixed random seeds, and reproduces results through R while flagging risks without certifying the analysis.
Create a skill from an idea or improve an existing draft through test prompts, output review, and benchmarks that compare results. The workflow can also refine a skill description to help it trigger for the right requests. A later part of the source material was unavailable.