hyperb1iss/hyperskills

آخر نشاط مصدر مسجل: ٢١ أكتوبر ٢٠٢٦ في ٢٢:١٠

آخر تحديث لفهرس SkillsMP: ٩ نوفمبر ٢٠٢٦ في ١٤:٢٢

skills مجمعة: ١٣

نجوم GitHub: ٣٦
تفرعات GitHub: ٢

التثبيت بمساعدة مساعد ذكي

انسخ هذا الطلب إلى المساعد الذكي الذي تستخدمه.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/hyperb1iss/hyperskills.

ملف README للمستودع

مصدر ملف README للمستودع: README.md

قراءة الملف الأصلي

عرض ملف README كاملًا

<h1 align="center"> ⚡ hyperskills </h1> Focused skills for things capable agents still need to learn  <img src="https://img.shields.io/badge/Skills-14-e135ff?style=for-the-badge&logo=anthropic&logoColor=white" alt="14 Skills">  <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-ff6ac1?style=for-the-badge" alt="MIT License"></a> ## What earns a skill A model does not need a tutorial for every tool it touches. Hyperskills carries the parts that are easy to miss: how to preserve another worker's edits during Git surgery, distinguish a review finding from a plausible story, verify a generated artifact at its destination, or extract useful memory without turning an old transcript into new instructions. The process skills help with decisions and evidence. They adapt to the task instead of prescribing agent counts, approval rituals, or a fixed pipeline. Domain skills remain where the operational details justify them: Git and terminal UI design. Use the installed tool's help and current official documentation for ordinary package management, linting, and type checking. Skills load independently. [Sibyl](https://github.com/hyperb1iss/sibyl) adds durable project memory when available. Host permissions and the user's actual request govern execution; a skill does not create authorization or override a settled decision. ## Installation ### Claude Code ```text /plugin marketplace add hyperb1iss/hyperskills /plugin install hyperskills@hyperb1iss ``` ### Skills CLI ```bash npx skills add hyperb1iss/hyperskills --all npx skills add hyperb1iss/hyperskills --skill implement ``` Use the installer's supported agent selection to choose your host. Check its current help rather than assuming every host has the same skill directory or invocation syntax. ### Local development ```bash git clone https://github.com/hyperb1iss/hyperskills.git cd hyperskills make install ``` The installer links each skill into `~/.agents/skills` and `~/.claude/skills`. Existing directories are backed up before replacement. Installing a plugin or updating this checkout does not automatically remove older standalone copies of retired skills; inspect those copies before removing them. ## Composing skills Start where the uncertainty is. A clear bug can go straight to implementation; an open product question benefits from exploration before a plan. The combinations below are examples. | Situation | Useful combination | | -------------------------------------- | ----------------------------------------------------------- | | Unresolved feature direction | brainstorm, then research where facts decide the choice | | Large change with known requirements | plan, implement, independent review | | Independent workstreams | orchestrate with explicit ownership and integration | | PR review | hyper-pr-review; cross-model-review for another perspective | | PR authoring | super-good-pr | | Prose cleanup | deslop with the artifact's existing voice and format | | Raster asset from another harness | codex-imagegen | | Durable lessons from past sessions | dream | | A product video that tests the product | demo-reel; cross-model-review for its verifiers | ## Process skills ### `brainstorm` Explore the problem and alternatives before committing to a design. Treats the user's hunch as a hypothesis, separates explicit requirements from inferred preferences, and identifies the experiment that would change the recommendation. Clear implementation requests do not need a brainstorming detour. ```text /hyperskills:brainstorm ``` ### `research` Gather evidence for a question, from a targeted lookup to independent research lanes. Tracks source lineage, contradictions, and benchmark comparability. Stops when the remaining uncertainty would not change the action, and makes the call plainly when the evidence supports one. ```text /hyperskills:research ``` ### `plan` Turn requirements into dependency-aware tasks with observable completion criteria. Keep tests with the behavior they verify, preserve deliberately open decisions, and give each worker enough context to resume without reconstructing the conversation. ```text /hyperskills:plan ``` ### `implement` Make a focused change and verify the behavior where another component consumes it. Covers error diagnosis, shared-workspace ownership, meaningful test receipts, compatibility, and recovery. Verification grows with the uncertainty and consequences of the change. ```text /hyperskills:implement ``` ### `orchestrate` Partition independent work, define ownership beyond file paths, and integrate results against the original request. Includes worker and reviewer briefs, lifecycle tracking, artifact verification, and honest handling of incomplete work. Agent count follows useful parallelism and available capacity. ```text /hyperskills:orchestrate ``` ### `cross-model-review` Launch and consume an independent review of code, a specification, a diagnosis, or factual claims. Covers current CLI discovery, process handles, output capture, permissions, billing intent, and evidence-based finding disposition. An optional Claude runner exposes recent activity, retains private logs, and checks the declared Git scope for changes. A second model adds a perspective; agreement still needs proof. ```text /hyperskills:cross-model-review ``` ### `hyper-pr-review` Conduct the review yourself. Trace changed behavior and actual callers, try to disprove each candidate, and explain each finding as a concrete cause and consequence. Group repeated symptoms under their shared remedy and distinguish required changes from useful options. Deep review adds deliberate coverage of rollout, security, and structural quality. Unfinished required checks stay visible beside the verdict and prevent approval. ```text /hyperskills:hyper-pr-review ``` ### `codex-imagegen` Delegate raster generation or editing when another harness needs Codex's image capability. Preserve reference roles and edit invariants, keep jobs and receipts distinct, and verify the delivered files. Prefer native image tools when already available; capability and billing depend on the configured environment. ```text /hyperskills:codex-imagegen ``` ### `super-good-pr` Write and maintain PR descriptions with a clear title, a causal opening, and a visible path through the mechanism and evidence. Worked examples show how a small fix, a complex boundary, and a scoped update need different amounts of explanation. Preserve human-authored sections and template requirements. Refresh claims when the artifact changes, and distinguish a rewritten commit identity from changed content. ```text /hyperskills:super-good-pr ``` ### `deslop` Edit prose while preserving meaning, uncertainty, and the author's voice. Includes a protection-aware scanner, surface profiles, editing techniques, and a dated evidence catalog. Scanner candidates prompt judgment; stylistic patterns do not establish authorship. ```text /hyperskills:deslop ``` ### `dream` Review authorized conversation history for decisions and recurring lessons that should survive a session. Separates visible evidence from speculation, preserves provenance, checks current state before storing claims, and keeps project scope and sensitive data boundaries intact. ```text /hyperskills:dream ``` ### `demo-reel` Make a demo reel that doubles as an acceptance test of the real product. Every shot in the list is a test, a failed take starts a fix lane with evidence, and the whole reel is reshot until it passes. Bundled scripts check the toolchain, capture web pages deterministically at 4K, film an agent with a throwaway HOME, a dedicated login, and an allowlist (brakes rather than isolation, with a proof that shows which ones hold), and verify rendered video. An agent handed the skill's URL can clone it and run them; ffmpeg is the only system dependency, and browser capture adds Node and Playwright. ```text /hyperskills:demo-reel ``` ## Domain skills ### `git` Handle rebases, stacked branches, conflict resolution, and shared worktrees. Preserve inspected remote expectations with explicit leases, compare rewrites using the appropriate evidence, and commit only owned staged content. Recovery begins by preserving the candidate state. ```text /hyperskills:git ``` ### `tui-design` Design terminal interfaces around capabilities, state, input, and lifecycle. Covers layout and visual hierarchy alongside Unicode width, fallback rendering, keyboard access, resize behavior, and terminal restoration. Includes a visual catalog and examples from existing applications. ```text /hyperskills:tui-design ``` ## Maintaining the library Metadata supports discovery; the entrypoint contains the essential decisions; references hold conditional detail. Short is useful when the procedure remains complete. A long reference is worthwhile only when an agent can find and use the relevant part. Version 3.11 removes the `uv`, `ruff`, `ty`, and `uv-build` skills. Their generic tool guidance belongs in model knowledge and current official documentation, while copied version tables require maintenance without adding enough value. Version 3.12 also retires the Tilt skill. Use current Tilt documentation and keep project-specific development patterns with the project. The [September 2026 review](docs/library-review-2026-09.md) records the substantive changes and research limits. The [evaluation cases](evals/README.md) exercise task behavior and routing; schema validation alone cannot establish that a skill helps. ## Development The checks use `uv` (Python and isolated PyYAML), Bash, Perl, and ffmpeg (for the demo-reel selftest). Markdown linting and formatting use the tools named in the Makefile. ```bash make check # Parse manifests, YAML frontmatter, and bundled references make test # Run validator, scanner, review-runner, and demo-reel regressions make all # Run both checks and tests make lint # Optional JSON, YAML, and Markdown lint tools make format # Format tracked Markdown and JSON make sync-metadata # Regenerate the skill badge, inventory, and repository tree ``` The validator checks concrete `references/`, `scripts/`, and `assets/` declarations in skill entrypoints. It does not execute examples, crawl external links, or grade agent behavior. Review reference contents and run relevant behavioral cases when changing a workflow. CI also checks derived metadata and release consistency. The [release workflow](docs/releases.md) prepares a single PR containing the version, changelog, and release notes, then publishes the exact merged commit after validation. Ordinary skill changes leave the version unchanged until release preparation. See [AGENTS.md](AGENTS.md) for contribution guidance. Licensed under [MIT](LICENSE). 💜 Built by <a href="https://hyperbliss.tech">Hyperbliss Technologies</a>

جارٍ تحميل ملف README للمستودع…

Skills في هذا المستودع

متخصصو إدارة المشاريع مطوّرو البرمجيات الكتّاب التقنيون المهن الحاسوبية الأخرى

٧ فئات مهنية · 85% مصنفة