Find the open-source solution that fits.

AI Tutor structures learning around retrieval practice, spaced review, interleaving, elaborative questions, and adaptive difficulty rather than immediate answer delivery. Its learn, practice, review, quiz, and explain modes prompt the learner to recall first and offer only the hint needed to continue. It is plain Markdown that can be adapted to several LLM platforms; where file uploads are supported, the learning-science reference files can be added as well.

decide researches current industry practice for an explicit implementation question, then acts autonomously when evidence converges on a reversible choice. When sources disagree, it asks for a factual piece of project context instead of presenting an arbitrary preference poll. This repository is now a legacy mirror whose active development moved to Decision Kernel, and its workflow depends on external research tools.

oss-due-diligence guides a four-phase assessment of open-source repositories and third-party dependencies: identity and license checks, adversarial review, operating posture, and incident response. It adjusts emphasis for project archetypes and includes shell scripts for dependency hygiene, risky patterns, secrets, and release signals. The methodology is explicit about what was checked and avoids treating a limited review as proof that a dependency is safe.

perplexity-plugin teaches Claude Desktop and Cowork how to route fresh-information tasks across Perplexity's search, answer, reasoning, research, computation, history, and diagnostic tools. It also handles plan-gated features by falling back to free-tier search patterns when needed. The package does not install an MCP server: users must connect one themselves, and plan limits should be checked as they change.

compass audits a long Claude Code session along two axes: drift from the original task and rot in the codebase. It reads the transcript plus cached Git, test, lint, file-size, TODO, and boundary signals, then returns a SAFE, CONSIDER, or STOP recommendation with a next action. It does not run tests or research the web, supports Claude Code JSONL transcripts only, and is now a legacy mirror.

Quantitative Researcher supports audit-and-reproduce, analyze, critique, and consult modes for research data and papers. It inspects coding and assumptions, flags common errors such as suppression or causal overreach, and performs calculations with Python statistics libraries before writing APA-structured methods, results, and limitations. The skill depends on an execution-capable environment; it is not designed to replace computation with model reasoning.

mj-agent expands one Midjourney prompt into ten raw variations or a single brace permutation that can be pasted into Discord for batch generation. It can also split oversized commands and compare prompt runs across model versions. The number of permutation jobs depends on the Midjourney plan, permutations require Fast mode, and the skill prepares prompts rather than creating images itself.

Journalism Skill helps journalists and researchers organize investigations around retrievable sources, careful causal language, and explicit verification. Its guidance covers public-records work, interview planning, data review, sensitive allegations, and right-of-reply handling, with reusable templates and examples for both agentic and web-chat settings. It structures the work but cannot turn unsupported claims into verified facts.

curiosity-merge is a separate sharing and federation layer for curiosity-engine wikis. It exports self-contained subgraphs, discovers and accepts possible cross-page bridges, merges by stable IRI with provenance and collision handling, and reverses merges through manifests. Missing source bytes can be marked and later rehydrated with the receiver's own access.

This skill gives AI coding agents a worktree-based workflow for parallel feature branches. It initializes projects, uses Portless named local URLs, isolates local state, and supports safe resume, browser review, test-data cleanup, and branch finishing. It can also check whether independent local databases are available before parallel stateful work.

korean-privacy-terms is a Claude skill that uses a short interview about service type, users, regions, and operator location to choose a jurisdictional path. It produces Korean, English, or bilingual privacy policies and terms with templates for Korea's PIPA and the EU GDPR. The project clearly presents the output as a draft for legal review.

This repository currently packages a chemistry-toolkit skill for Claude Code and the Claude Agent SDK. It can render molecular structures with RDKit and 3Dmol.js, typeset equations with KaTeX and mhchem, balance and verify reactions, embed editable structure sketchers, and produce step-by-step problem layouts. The repo also documents symlink installation and .skill bundling.

harness-engineering-skill bootstraps a project harness for reliable LLM-agent work and maintains it across sessions. Its four modes cover project setup, handoff, promoting repeated mistakes into stronger guardrails, and independent evaluation. Deterministic shell scripts handle repeatable operations, while Markdown modes leave judgment to the agent; project artifacts record configuration, progress, features, lessons, and evaluation reports.

This Claude skill guides short- and long-form novel creation through 13 Snowflake Method steps, from theme and one-sentence premise to characters, layered outlines, scene planning, chapter generation, and prose conversion. Its final pass scans for technical wording, awkward logic, and stiff tone, then applies five style adjustments and shows the original alongside the revised text.

architecture-solver turns workload requirements into infrastructure designs through seven dimensions: TIME, SPACE, WORK, STATE, FAILURE, COST, and TRUST. It applies formulas such as Little's Law, Erlang-C, USL, availability calculus, and cost optimization, then matches technologies and proven patterns. The skill can produce sized and costed recommendations, compile them to Terraform, CDK, or Pulumi, and define load, resilience, security, and cost checks.

mailbox-cleanup combines a Python CLI with a Claude Code skill for IONOS IMAP mailbox triage. It scans messages, previews destructive changes, and defaults to soft-delete or other dry-run operations. Filters support deletion to Trash, moving, archiving, unsubscribe handling, exact Message-ID deduplication, bounce detection, and large-attachment listing. Applied actions are appended to an audit log, while permanent deletion remains manual.

This Databricks skill turns an EHR-derived Unity Catalog bronze layer into an OMOP CDM v5.4 Asset Bundle project. It generates per-table YAML transform configurations through a conversational flow, validates them with Pydantic, scaffolds jobs and data-processing files, and runs five OHDSI checks for schema, primary keys, concept foreign keys, domain conformance, and NOT NULL completeness. It validates 20 tables and auto-builds templates for 14.

servicetitan-forms-skill helps Claude build, fix, and refactor ServiceTitan Form 2.0 JSON. It covers the envelope and IsForm20 flag, conditional visibility and micro-rule splitting, common import validation failures, fieldType pitfalls, preview errors, and plain-English-to-JSON form generation. The repository says its patterns were checked against live tenant exports and that the skill ships only Markdown, without scripts or network calls.