Start with the job, then find an open-source answer

Reel Check downloads and transcribes Instagram, TikTok, or YouTube Shorts locally, extracts the video’s concrete claims, and checks named projects before judging the idea. It also compares the concept with your existing skills, automations, and configuration, then returns an already-have, implement, partial, skip, or interesting-only verdict. It never implements the recommendation automatically.

Mac Vibe Triage investigates memory pressure, battery drain, orphaned development processes, and browser helpers left behind by AI-assisted work. It gathers read-only evidence across CPU, cumulative time, RSS, process ancestry, settings, and installation locations before classifying candidate PIDs. A kill requires your explicit approval in the current session, and the safety checks reject root, system, other-user, and active-agent processes.

Hangeul DeAI edits Korean prose for translation-like phrasing, excess commas, repetitive endings, buzzwords, and a mismatch between the writer, document, and audience. It favors targeted edits over wholesale rewriting and checks that numbers, proper names, and claims remain unchanged. The skill is for natural, trustworthy writing, not for bypassing AI detectors or adding artificial mistakes.

Draw.io Diagrammer generates flowcharts, SOPs, ERDs, architecture diagrams, UML class diagrams, and sequence diagrams from reusable XML templates. Every render goes through a screenshot, inspect, fix, and re-export loop to catch overflow, cramped spacing, collisions, and bad arrow routing. It requires the draw.io desktop binary; Linux or headless runs also need xvfb.

PDF Bookmarks reads a printed table of contents from scanned or OCR’d books and writes matching PDF bookmarks and page labels. It corrects front-matter offsets between printed and PDF pages, and can fall back to font-based detection or a readable manual list when parsing is uncertain. OCR options include Tesseract, macOS Vision, and Windows OCR, but the Windows backend is still beta and unverified on real Windows hardware.

Social Media Content Planner takes a client from onboarding to a dated monthly calendar. It collects brand and audience constraints, organizes research and content in a Google Spreadsheet, and researches current topics through web search, with the YouTube Data API available as an optional enhancement. Explicit approval loops run in Slack or chat before the calendar is handed to production.

Startup Kickoff puts a startup idea through twelve evidence-based validation gates and summarizes the result with a Survival Index and an Autopsy Card or, for ideas that clear the gates, a Survival Certificate. Weaker ideas can continue into the Reforge process to develop pivot candidates. The output is meant to expose assumptions and direct further evidence gathering before a team commits to building, not replace real market validation.

BlueNexus Agent Builder interviews you about an agent’s audience, workflow, guardrails, connectors, knowledge, and dashboard before driving the BlueNexus studio through Claude in Chrome. It configures the agent and publishes it end to end, with room to refine the dashboard from reference screenshots. Automated building requires a logged-in BlueNexus account and Claude in Chrome; without browser control, the skill serves as a step-by-step guide.

Coverbean turns a post description into a title, image prompt, and platform-sized cover or thumbnail for Xiaohongshu, WeChat, Douyin, TikTok, and YouTube. It renders the image through apimodels.app’s gpt-image-2-all and crops it to the required aspect ratio, with optional ffmpeg support. Setup needs an API key and Node.js 18 or newer, and image generation is paid.

Gatefrisk reviews HTTP routes touched by a change, classifies them as read-only or mutating, and blocks completion when a route can write data, move money, or trigger a job without authentication. It is designed to catch temporary trigger endpoints and similar authorization gaps before release. The skill supports Claude Code, Codex, Cursor, Gemini CLI, and OpenCode.

Hookfrisk inspects agent hooks changed by a diff and looks for untrusted tool output, filenames, or trigger text being piped into a shell, along with patterns such as curl piped to bash. It refuses the done state while an auto-running hook can execute input unsafely, helping catch command-injection and remote-code-execution paths before release. It is intended for Claude Code and compatible coding agents.

Revenue Messaging Framework audits a B2B homepage, scores its messaging, explains weaknesses section by section, and returns rewritten headlines, value propositions, calls to action, and a three-layer narrative. It is meant to clarify what a go-to-market team should change on its own site, combining an open methodology with a Claude Code skill. The output is an audit and rewrite, not a general page builder.

CruiseFeed Skill teaches an agent to query normalized live cruise data across more than 60 cruise lines. It selects the right endpoint, filters sailings and fares, handles pagination, and reads only fields returned by the API for questions about dates, destinations, ships, prices, and ports. It is a query guide for the CruiseFeed API rather than a booking system.

SEO Launch Resonance organizes an organic-search launch around three inputs: an indexing channel, targeted landing content, and real demand. It guides an agent through search-console setup, internally linked pages without cannibalization, dated supporting content, and per-page attribution or bounce-rate checks. The workflow needs site and deployment access plus analytics accounts, and its playbook is not a guarantee of traffic growth.

Webcam Snap lets Claude Code capture a still image from a Linux V4L2 webcam with ffmpeg, skipping initial frames so auto-exposure can settle. The resulting photo can be described or analyzed by a multimodal model, or handed back to you. It works as a skill, plugin, or raw command, but requires a readable /dev/video device and ffmpeg.

Stratos Hook Writer mines an existing post, article, script, transcript, newsletter, case study, or idea for its strongest concrete asset, then creates five distinct hooks tailored to the selected platform and commits to one recommendation. Its playbooks account for mechanics such as the LinkedIn fold, TikTok’s layered opening, YouTube’s first seconds, and email subject lines. The hooks are adapted to the source rather than filled from a generic template.

Spine turns disposable chat into an append-only planning record. When a thought settles, /spine writes a compact, typed node into drafts/spine, and a later session can resume from that file instead of replaying the conversation. Each node type has its own gate: factual nodes need a citation, decisions need a reason and rejected alternative, definitions record the old meaning, and open nodes state what blocks resolution. The license permits noncommercial use; commercial use requires a separate agreement.

PPT-to-MD extracts text, speaker notes, and raw data from native charts in .pptx files using a deterministic Python script. It can emit Markdown for Obsidian, Mermaid bar charts, and optional YAML frontmatter without OCR or machine-learning dependencies. The tool only accepts .pptx, and charts pasted as images cannot yield their underlying numbers.