待翻译：Vidbyte Skills – Skills That Keep AI-Assisted Work from Becoming Passive

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Notifications You must be signed in to change notification settings Fork 0 Star 0 BranchesTags Open more actions menu Latest commit History 210 Commits 210 Commits Folders and files NameName Last commit message Last commit date agent-skills/core agent-skills/core artifacts artifacts bin bin cli cli docs/design docs/design lib lib packages packages scripts scripts skills skills .env.example .env.example .gitignore .gitignore LICENSE LICENSE README.md README.md llms.txt llms.txt package.json package.json skills-manifest.json skills-manifest.json Repository files navigation Vidbyte helps developers package reusable agent workflows, learning routines, and reasoning methods as portable skills. Vidbyte Skills is the repository and npm package that installs those skills into local coding harnesses and also ships the vidbyte command used by skills to submit authenticated artifacts back to Vidbyte. Coding agents need local, versioned instructions that are more precise than a general prompt — Vidbyte Skills packages those instructions as auditable Markdown folders, installs them into the harnesses developers already use, and keeps the authenticated network boundary in code rather than prompt text. The repository owns the skill source files under skills/; the installer copies or links those skills into native skill directories and writes generated rule files for Claude Code, Codex, Gemini CLI, OpenCode, Cursor, Hermes, GitHub Copilot, Warp, Factory, Crush, Aider, Augment, Cline, Continue, Roo Code, Windsurf, and AGENTS.md-compatible tools. Repository: https://github.com/cerredz/Vidbyte-Skills Repository Map Area Role skills/ Source of truth for installable skills, catalog categories, authoring rules, and validation expectations cli/commands/ Python command layer for auth, feedback, compression, retention, dry-run validation, and backend submission lib/ Node installer internals for skill discovery, catalog filtering, target resolution, and install actions bin/ Package binary shims for vidbyte, vidbyte-skills, category installers, and roleplay installers scripts/ Validation, smoke testing, packaging, and catalog-generation scripts Intended Behavior Run the installer from npx, npm scripts, or the package binary. By default it installs the curated version 1 skills into user-level harness directories on the local machine. Use --version all to install every valid skill from this repository, or name one or more skills to install only those skills. After installation, open or reload your coding harness. The skills are available from that harness according to its normal local skill or rule discovery behavior. Install Install the default skill set: npx vidbyte-skills Install one skill by name: npx vidbyte-skills Updating Skills Run the update command to pull the latest published version of vidbyte-skills from npm and reinstall all skills into your harnesses automatically: vidbyte-skills update Or via npx without a local install: npx vidbyte-skills update The update command: Reads your current installed version from package.json. Fetches the latest version from the npm registry. Exits cleanly with "Already up to date" if no new version is available. Runs npm install --global vidbyte-skills@latest if a newer version exists. Reinstalls all skills into your harnesses using the installer defaults (user scope, all platforms, version 1). Vidbyte CLI This package also exposes a vidbyte command for skill-to-backend submissions. Skills should call the CLI instead of constructing backend requests directly in prompt text. vidbyte feedback submit --file feedback-log-2026-05-11-example.md --domain software-engineering --conversation-id example The /retain skill uses an argument-shaped command so the model can pass the generated exercise fields directly: vidbyte retain --title "Retain this session" --domain software-engineering --conversation-id example --concept1-name "CLI auth boundary" --concept1-distillation "The prompt generates content while the CLI owns authenticated transport." --concept1-anchor "A locked dispatch box with a public label and a short-lived pass inside." --concept1-hook "This matches short-lived deployment credentials, where the long-lived account session only mints scoped one-use credentials." --question1 "Why should the prompt not construct Vidbyte auth headers itself?" --answer1 "A strong answer says prompt text is not a trust boundary, while the CLI keeps account sessions and requests short-lived invocation tokens in code." The CLI is implemented in Python (stdlib only, no dependencies). It authenticates with a Vidbyte session, mints short-lived invocation tokens for backend writes, sends traffic only to https://vidbyte.pro, and never asks prompts to construct auth headers. Copy .env.example to .env for local development; real .env files are ignored by git. Use --dry-run to validate command input without sending a network request: vidbyte feedback submit --file feedback-log.md --domain software-engineering --conversation-id local-test --dry-run vidbyte retain --concept1-name "Concept" --concept1-distillation "Mechanism" --concept1-anchor "Vivid image" --concept1-hook "Personal hook" --question1 "Question?" --answer1 "Answer key" --dry-run The CLI can also be invoked directly via Python: python3 -m cli feedback submit --file feedback-log.md --domain software-engineering --conversation-id local-test --dry-run python3 -m cli retain --concept1-name "Concept" --concept1-distillation "Mechanism" --concept1-anchor "Vivid image" --concept1-hook "Personal hook" --question1 "Question?" --answer1 "Answer key" --dry-run Agent Usage Guides The vidbyte command also serves bundled agent-facing usage guides so coding agents can load instructions that match the installed CLI version. vidbyte agents vidbyte agents get core vidbyte agents get core --full vidbyte agents get core --json vidbyte agents only serves CLI usage instructions. Use vidbyte-skills to install product skills into local coding harnesses. For development and tests, set VIDBYTE_AGENT_SKILLS_DIR to point at an alternate agent guide directory. Skills Learning Skills for active learning, comprehension, retention, and research. Skill Command Description 1-3-5 /1-3-5 Plans exactly one Big, three Medium, and five Small daily tasks and logs every displacement instead of allowing a tenth blindspots /blindspots Surfaces hidden principles, tradeoffs, or considerations the user hasn't named yet using targeted guiding questions bullet-journal /bullet-journal Sets up Ryder Carroll's four analog modules, Rapid Logging syntax, daily logs, and mandatory monthly Migration chain-method /chain-method Builds vivid item-to-item links for exact forward recall of short ordered lists; warns when length or random access needs another method compression-check background Silent background coach that asks you to explain what you just built; evaluates your response and submits a learning record to Vidbyte concept-coverage /concept-coverage Background tracker that monitors how deeply you engage with a concept; export as a JSON learning artifact at end of session daily-review /daily-review Extracts high-risk session concepts at end of a work session, appends them to a log, and sends them to Vidbyte for spaced review dominic-system /dominic-system Builds an initial-based 00–99 Person/Action system for long numbers and card sequences, then scores exact decoding drills dr-ta /dr-ta Guides testable prediction, reading, evidence verification, revision, and final prediction-evolution synthesis explain /explain Rebuilds explanations from first principles; diagnoses which understanding layer is broken and fills from the lowest solid floor explain-away-others /explain-away-others Before accepting your approach, identifies 2–3 competitive alternatives and requires mechanism-level explanations for why each fails feedback-generator /feedback Multi-agent harness that generates expert domain feedback grounded in 180+ learning-science papers through iterative self-refinement find-papers /find-papers Searches for academic papers via plain-language prompt, filters by credible databases, and returns a clean formatted list finding-resources /finding-resource Produces a comprehensive learning-resource map across books, papers, courses, and practitioner writing for any topic flow-notes /flow-notes Captures argument structure as big-idea phrases and arrows, then requires a throughline synthesis and deep connections gtd /gtd Runs David Allen's Capture, Clarify, Organize, Reflect, Engage workflow with physical Next Actions and Weekly Review insert /insert Applies five active-reading symbols, builds review lists, and verifies every "already knew" confidence mark interactive /interactive Routes the current conversation to SQ3R, PQ4R, or PAO and starts the selected gated workflow jargon /jargon Surfaces domain-specific jargon, translates to plain language, and builds vocabulary before engaging a technical topic learn-from-video /learn-from-video Browser-controlled active learning session for a YouTube video with transcript-based segment planning and checkpoint questions misconceptions /misconceptions Silently tracks faulty mental models during the session and writes an end-of-session misconception log motivate /motivate Delivers one non-repeated motivational learning quote and logs it so it is never shown again my-knowledge /my-knowledge Scans the session to give an honest assessment of genuine understanding vs. context-dependent familiarity ok5r /ok5r Guides long structured reading through Survey, Question, Read, Record, Recite, Review, and Reflect to produce a reusable study sheet pao-system /pao-system Builds and drills a persistent 00–99 Person-Action-Object memory system for long ordered sequences para /para Classifies digital notes and files into Projects, Areas, Resources, or Archives by actionability rather than topic pq4r /pq4r Guides dense structured reading through Preview, Question, Read, Reflect, Recite, and Review practice /practice Creates high-volume practice questions that emphasize pattern recognition, variation, and creative intelligence question /question Produces detailed five-section answers (What, Why, Critical Thinking, Best Practices, Resources) to counter shallow responses in coding harnesses question-builder /question-builder Background tracker that logs retention and future-direction questions throughout the session; export as a JSON artifact read-paper /read-paper Reads a research paper (arXiv, DOI, PDF, Semantic Scholar, PubMed), strips noise, extracts a 6-field core signal, and runs a learning gate check reap /reap Guides Read, Encode, Annotate, and Ponder for sources where accurately representing the author's argument matters more than fact extraction research /research Answers grounded in verified knowledge with explicit source attribution and epistemic labeling on every claim; peer-reviewed sources only retain /retain Pauses the conversation, generates a 15-minute retention exercise from the session, and submits it to Vidbyte scope /scope Defines the boundaries of broad domains, highlighting core, adjacent, and commonly misattributed fields solo /solo Classifies an unaided explanation from Pre-structural through Extended Abstract with evidence and a next study action sq3r /sq3r Guides structured nonfiction through five active reading phases with paraphrase and recitation gates struggle /struggle Background tracker that logs repeated struggle patterns and blind-spot signals throughout the session; export as a JSON artifact theoretical-feedback /theoretical-feedback Extracts the underlying mental model separating novices from experts for any situation or mistake in any domain think-aloud /think- [truncated for AI cost control]