Find the open-source solution that fits.

Anti-Apophenia makes an agent treat an appealing pattern as a hypothesis rather than a finding. It requires a claim, evidence, null model, alternatives, prediction, and disconfirmation loop, while penalizing post-hoc fitting, repeated attempts, cherry-picking, and vague resemblance. Evidence grades and explicit verdicts separate strong, weak, rejected, and unknown results, and tempting branches stay parked instead of steering the investigation. Speculative exploration is allowed only in a clearly marked sandbox.

hekouwang-stock-data-reader-skill accepts an A-share code, pulls public market and company data through AkShare, calculates a fixed set of indicators, and produces publishable chart cards plus long-form and headline versions. Its built-in boundary is deliberately non-predictive: it does not recommend stocks, forecast prices, or give trading points, and adds a risk notice to the output. The data source needs no API key, but the result remains an informational retrospective rather than investment advice.

github-auto-backup interviews you about repository settings when creating a project, then manages local Git and GitHub backups with conventional commits. It can commit, push, pull, and create branches, but pauses before public pushes, force-pushes, opening or merging pull requests, deleting branches, or changing repository settings. It never commits secrets or silently makes a private repository public. Local-only mode works without GitHub; GitHub operations need an authenticated gh CLI.

Living Docs treats documentation as a governed part of the codebase, organizing constitutions, PRDs, ADRs, BDRs, issues, research notes, and Mermaid diagrams around five invariants. Facts have one home, every document is indexed, old decisions are superseded rather than rewritten, and structural changes carry their documentation in the same change. A deterministic linter checks the contract across Claude Code, Cursor, Copilot, OpenCode, Codex, and Pi. It governs organization, not a project’s technology choice.

The skill assigns each request an S, A, B, or C level based on complexity and risk, then scales the planning depth accordingly. For larger jobs it breaks down goals and constraints, audits available agents and tools, evaluates feasibility, and lays out milestones and deliverables before implementation begins. Code is held until the user approves the plan, and the supplied planning template can be adapted to a team's workflow.

Recite Coach uses an agent workflow to turn PDFs, lecture notes, textbooks, tables, or pasted text into compact recall cards, then converts cards.md into cards.json for a local browser player. The player supports answering, revealing, self-rating, and Leitner-style review on the user’s machine, with no cloud study service required. The skill creates the cards; the player is not a standalone PDF-understanding engine. Python’s standard library is sufficient for the local tool.

Breakglass helps Solana teams respond to failed transactions, suspicious drains, authority compromise, RPC degradation, and frontend incidents without rushing into risky actions. It organizes evidence, labels severity and confidence, then routes the case to focused debugging, communication, recovery-planning, or postmortem guidance. The skill never requests keys, submits transactions, moves funds, or changes authorities; humans remain responsible for every on-chain action.

Automatically route a hard task into either a creative track that amplifies, builds, playtests, and judges the artifact, or an audit track that gathers independent findings, reaches consensus, implements fixes, and verifies them adversarially. The Claude Code skill loops until explicit completion criteria are met and a fresh pass finds no material next steps, escalating with preserved state when blocked or over budget.

Use a resume and portfolio to collect active Korean job listings from eight platforms, rank them with stated matching criteria, and explain both fit and gaps. The Claude Code skill then develops role-specific Korean jasoseo strategies from the candidate’s own material rather than inventing experience. Some sources expose only listing-level details, and unofficial platform endpoints may need maintenance when they change.

PDF Email Optimizer converts common Office files when needed, then reduces the resulting PDF toward a requested attachment limit through structural cleanup and controlled image recompression. It keeps the source untouched, reports the chosen strategy and quality trade-offs, and can render before-and-after pages for visual checks. Office inputs require LibreOffice, while unusually tight size targets may require an explicit lossy compression mode.

durable-loop adds persisted checkpoints, crash recovery, idempotent retries, append-only logs, and machine-checked completion to Claude Code’s built-in /loop. A verify_done rule requires consecutive passes before a run is considered converged, and reusable learnings can carry across runs. It is designed for long tasks that may outlive a process, while stricter dangerous-operation brakes are opt-in rather than enabled by default. The scripts use Python’s standard library.

FiveM Resource Builder scaffolds resources such as shops, garages, jobs, and HUDs for FiveM or RedM while putting security-sensitive logic on the server. Generated events validate types and ranges, prices stay server-controlled, money and item actions get rate limits and anti-duplication guards, and world actions use proximity checks. It supports ESX, QBCore, QBox, ox, and related ecosystems, with escaped NUI output and optional luacheck CI. The result is a secure starting point, not a finished game feature.

Upload PDF, spreadsheet, or screenshot reports from Russian brokers to consolidate positions across accounts. The skill uses current MOEX, Bank of Russia, and cbonds data to assess duration, after-tax yield, concentration, and possible replacement candidates, with optional Monte Carlo risk scenarios. It focuses on Russian bonds and equities, does not execute trades, and presents analysis rather than personalized investment advice.

Data Animation Skills turns datasets into moving charts, animated infographics, or narrated presentation videos while keeping exact values visible in every frame. The pack supports bar-chart races, line and area reveals, count-up tickers, designed statistic scenes, and slide builds timed to voiceover, with batch rendering across multiple data files. It runs as agent skills for Claude Code, Cursor, Codex, and other compatible clients and is intended for analysts, product teams, and explainers.

Ad Video Skills helps marketing teams turn one motion-graphics template and a data table into on-brand ad variants for A/B testing, launch films, or animated testimonials. It can render multiple placements and organize hooks, beat-synced reveals, review quotes, and calls to action into repeatable video production.

YouTube Video Skills creates captioned audiograms with moving waveforms from podcast clips or voiceovers, and builds reusable YouTube intro and outro plates. The intro can be a short branded sting, while the end screen leaves room for YouTube’s clickable subscribe and next-video elements. Templates can be reskinned with a channel’s name, logo, colors, and audio cue, making the pack useful for long-form channels that also publish square or vertical social clips.

Freelance motion designers and studios can turn a vague client request into a sign-off-ready brief, calculate a defensible quote, and define revision limits or change-order language. Separate skills also prepare platform-specific delivery specifications, quality checks, and documented brand-motion rules. The pack installs in Claude Code and other agent-skill-compatible tools such as Cursor and Codex.

Start with a research idea, an existing draft, or source materials and move through scoped literature collection, DOI verification, evidence-matched writing, adversarial review, and prose cleanup. The self-contained Claude Code and Codex skill can export LaTeX, Word, and PDF while keeping an auditable record of writing and review decisions. Word and PDF output require pandoc and, for PDF, a TeX engine.