m4bwav

Evergreen chart and graph skills: an evidence-based knowledge base of chart types (when to use, what each excels at, when not to, substitutes, perceptual evidence, notes) and sixteen render targets (Markdown with Mermaid or PlantUML, web with Vega-Lite, ECharts, Plotly, Chart.js and Observable Plot, static PNG/SVG via Python, editable PowerPoint, Excel and Google Sheets charts, Word and Google Docs pictures, D2, QuickChart URLs, terminal sparklines), a deterministic chooser and builder CLI that takes data by file reference to save tokens, and a curation skill that researches novel chart requests and grows the knowledge base so the next request is cheaper.

Plan AI video like a film crew: a director router, a shot-list and coverage planner, a script supervisor that diffs every shot card against character, prop, location and axis bibles, a compiler that turns generator-neutral shot cards into prompts for nine video models, and a take reviewer that routes failures to the cheapest fix.

Film craft skills for AI video: cinewright-script (logline, beats, scene turns, dialogue for generated voices), cinewright-design (turnaround sheets, clean references, prop descriptions, costume arc, color script), cinewright-sets (set plans, per-wall location strings, master plates, coverage and reverse angles, set light, plate checks), cinewright-movement (weight, one phrase per shot, fights, battles, hard subjects), cinewright-camera (lens, depth of field, exposure, shutter, aspect and framing, lighting, color temperature), cinewright-history (movement, era, genre, director and cinematographer style cards), cinewright-edit (assembly, rule of six, J and L cuts, pacing, trimming generations, cutting battles), cinewright-finish (grade order, color spaces and delivery, crop to the film's frame, day for night, crowd multiplication, upscale last) cinewright-sound (layers and stems, generated audio, battle sound, mixing to a loudness target) and cinewright-voice (voice sheets, designed or cloned voices locked by reference clips, drift checks). Needs the cinewright plugin.

Maintainer skill for the cinewright knowledge base: cinewright-curate adds, verifies and retires reference entries and refreshes video model cards in a repository checkout. Not needed to make films.

Self-maintaining, self-proving skills and knowledge: adaptive web research on four tracks (the subject, the skills, plugins, MCP servers and knowledge graphs built for it, how others use agents on the same goal, and how they test that the job was done), never-teach-twice learnings capture, plugin-owned codemaps, a portable AI-preferences profile, and an eval suite per skill (trigger, action, outcome, decoys, baseline) that passes on evidence outside the transcript, with a bounded tuning loop that researches the subject's own testing when a skill fails, and a worth check that warns while a skill is made or changed when it reads as mostly cost and decides KEEP, TRIM or CUT from the with-versus-without A/B. Skills say what they need (tools, packages, keys, servers, models, accounts, and access to databases, telemetry, cloud roles and APIs) and what for: on a new machine or agent harness, or when something is missing, the agent explains it, installs what is safe after saying so, hands the user exact steps for the rest, drafts least-privilege access requests without ever granting itself access, and records the recipe that worked and how each attempt went for that environment. An end-of-session wrap-up harvests the session transcript and keeps only what will save time or tokens or raise quality, routed to its narrowest home. Converts existing skills and makes new ones evergreen by default. Lives in one public git repository (https://github.com/m4bwav/evergreen-protocol): every install is a clone that pulls to stay current and, only if its owner said yes at install, sends improvements to the protocol, skills and scripts back as pull requests it opens itself (log entries go straight to master from a maintainer's clone), with email as the fallback route.

Nothing an AI session learned is lost when you change model, session, tool or vendor. Two tiers of plain-markdown knowledge: a per-project doc set (solutions with dead ends, decisions with reasons, plans, a 50-line HANDOFF, an index) committed with the repo or kept in an excluded folder when the repo cannot hold it, and a user tier in a private vault repository (profile, environments, cross-project lessons). A privacy split keeps people, politics and credentials out of any shared repo. Hooks orient each session, nudge capture once, and at session end sync the vault and push the project's docs (a pull request whenever the push is not certain). The vault asks once for a private remote or creates one. Every entry carries a recheck date: the index, the session-start line and a BM25 search over titles, aliases and bodies flag what is due, and recheck and verify check a stored fix against the current code before an agent acts on it. Typed links, a deterministic maintain command, and a reproducible benchmark. Installs in Claude Code, Cowork, Copilot and anything that reads Agent Skills. Follows the Evergreen Protocol in pointer mode.

Name any subject and get a scout for it: everscout researches where the subject lives online, verifies each community and feed, and writes a beat (watch list, entity vocabulary, question bank, searches, study outline); then it reads Reddit, Hacker News, Bluesky, Mastodon, Lemmy, Discourse forums, newsletters, YouTube, GitHub, arXiv and Hugging Face at each platform's polite pace, distils one paraphrased note per thread worth keeping, tallies entities into signals (new, rising, fading), writes radar-style trend reports and a sourced baseline study, and drafts respectful questions to makers that you approve and post by hand, paced by a ledger and bound by a researched conduct code, answers any question about a subject like a reporter on the beat, from its notes plus fresh research, with sources and confidence, and keeps statistics per beat (a metric catalog, an append-only series from its tallies, GDELT news volume and Wikipedia pageviews, a review that promotes and archives metrics, chart-ready CSV for chartwright).

Makes prose read as written by a person: removes AI writing tells (dashes, stock words, not-X-but-Y contrasts, staged setups, inflated significance, chatbot residue), keeps sentences readable, and checks the result with a bundled stdlib script instead of a reread. Evergreen: it re-researches the signs of AI writing on a schedule and carries the evals that prove it works.

Reads, writes and maintains markdown so it works as an Obsidian vault and still renders on GitHub and in VS Code: a lint for broken and ambiguous links, orphans, duplicate basenames, missing index files and bad frontmatter; an index (map of content) writer; a wikilink-to-relative-markdown-link converter; one command to set a folder up as a vault; and the official Obsidian CLI when it is enabled. Evergreen: the skill re-researches Obsidian, its CLI and the agent integrations around it on a schedule and carries the evals that prove it works.

Modernize an old published library end to end and release it again safely: survey, a golden capture of the published behaviour, a decisions-table plan that stops for the maintainer, a tested rewrite, an independent review, repository cleanup, a release rehearsal through trusted publishing with a human approval, registry verification and a handoff. npm and NuGet complete; PyPI, crates.io, Maven and Go documented and unverified. Nothing is published without the maintainer's approval. Evergreen.

Evergreen three.js skills for AI agents: current r186 APIs (WebGL, WebGPU, TSL), a knowledge base of topics, scenarios, libraries, recipes and rules with release ranges so stale advice stays labelled legacy, verified starters (single-file, WebGPU, Vite, React Three Fiber, 3D chart, video turntable, Rapier game), and a zero-dependency Node CLI that checks pages headlessly as text first (errors, scene numbers, pixel evidence), lints stale APIs against the installed release, reports glTF models, and records frame-exact video.

Evergreen skills for driving Unity with AI coding agents: the official Unity CLI and Pipeline package, headless batchmode and compile checks, MCP bridge selection, agent-safe Unity workflow rules, and UPM package management. Each skill re-researches its subject on a schedule and carries the evals that prove it triggers and acts.

Write and update a repository's GitHub wiki from the repository and its registry, with every example verified against the published package (npm and NuGet): survey, page set per repository kind, sidebar and footer, publish through the .wiki.git repository, live check, and a note on how to update it at the next release. Evergreen.