SylphxAI/skills

Profil SylphxAI

Dernière activité source enregistrée: 3 octobre 2026 à 02:31

Catalogue SkillsMP mis à jour: 17 octobre 2026 à 00:09

skills collectés: 60

Étoiles GitHub: 2
Forks GitHub: 1

README du dépôt

Fichier source du README : README.md

Lire le README source

Afficher le README complet

# Sylphx Agent Skills (`SylphxAI/skills`) Opinionated, scenario-driven Agent Skills and colocated reference methodologies for autonomous AI agents and human-agent teams across software engineering, system architecture, product design, commercial/partner negotiations, operations, marketing/growth, customer support, and clear communication. ## Core Design Philosophy: Opinionated Methodology vs. Organization State 1. **Skills Encode Chosen Methodology (Not Neutral Option Lists or State)**: - In every discipline—how we think and design, how we deliberate as a team, which technology stack we build on, how we score risk, how we ship software, and how we write—there are competing schools of thought. - Each skill in this repository states **the specific methodology we chose and how to execute it deterministically**: - **How We Think & Design (`self-challenge`)**: Simple Concept / Powerful & Deep Implementation (maximum value from minimum resources), Plan for 10x/100x Scale, Nothing Runs Always-On (scale-to-zero), Single Source of Truth (SSOT), SOTA Category-Leader Benchmarking, Opinionated Stack Selection (`stack.md`), and Quantitative `5×5` Risk Scoring (`risk.md`). - **How We Deliberate (`swarm-fuse`)**: Dynamic Group Facilitation (`Topic × Participant Context`), Independent Delphi + Red-Team Challenge Angles, and Single-Writer Coordination. - **How We Build, Unblock & Ship (`author-item`, `deliver-and-verify`, `prove-it`, `unblock`, `refactor`)**: Three-Stage Acceptance Test-Driven Development (ATDD), Work-in-progress limits and work item age, Release Scope Management, Upstream Contribution and Separation of Concerns (first-party client, zero backdoors), Complete Replacement + Systemic Corrective Action + Mistake-Proofing, and Primary-Source Verification. - **How We Communicate & Serve (`clear-writing`, `customer-support`)**: `Audience × Purpose × Channel` calibration, Concrete `❌ Bad vs ✅ Good` side-by-side examples, and Deterministic Support Case-State Machines. - **Organization State Lives Outside Skills**: Company-specific state (current headcount/size, active project portfolio, live server inventory, or specific rows in an accepted-risk register) belongs in the workspace repository or **Organization Knowledge**—never inside a skill. 2. **Three-Stage Acceptance Test-Driven Development (ATDD) for Every Outcome**: - **Stage 1 — Define Acceptance Test Cases First (`author-item` + `self-challenge`)**: Lock down observable positive and negative **Acceptance Test Cases** and pass the Category-Leader Benchmark, YAGNI (You Aren’t Gonna Need It), and Complete Migration tests before implementation starts. - **Stage 2 — Verify the Plan Passes the Test Cases (`self-challenge` + `swarm-fuse`)**: Verify that the proposed commercial, operational, growth, product, or technical plan satisfies every Stage 1 acceptance test case, carrying a same-session dated read of any external specification, law, or vendor API. - **Stage 3 — Execute and Verify at the Real Surface (`deliver-and-verify` + `prove-it`)**: Execute through one writer, complete caller migration and old-path deletion in the same delivery, and verify the delivered outcome at the real user, partner, customer, or operator surface. 3. **Role-Neutral Triggers & Colocated Reference Standards (Progressive Disclosure)**: - Skills trigger on the **observable task moment**, never on a static persona. Every deep domain standard lives in `skills/<name>/references/*.md` beside the skill that uses it, so agents load only the exact reference needed for the task at hand. ## Skills Catalog (10 Skills) | Skill | Trigger Moment | Colocated Reference Standards (`references/`) | | --- | --- | --- | | [`author-item`](skills/author-item/SKILL.md) | Writing or revising a work outcome item or ticket before execution (Stage 1 ATDD, CD3 priority) | `work.md`, `priority.md`, `quality-signals.md` | | [`self-challenge`](skills/self-challenge/SKILL.md) | Designing/comparing architectures or tech stacks, stress-testing plans/contracts/budgets/risks, questioning over-engineering, or checking root causes | `architecture.md`, `stack.md`, `commercial.md`, `growth.md`, `mechanics.md`, `mechanics-catalogue.md`, `games.md`, `product.md`, `experience.md`, `feedback.md`, `risk.md`, `conduct.md`, `agent-voice.md` | | [`deliver-and-verify`](skills/deliver-and-verify/SKILL.md) | Executing, finishing, merging, deploying, and verifying a change end-to-end (Limit Work in Progress, Release Scope Management, Upstream-First, Complete Replacement) | `code.md`, `auth.md`, `security.md`, `integration.md`, `identifiers.md`, `performance.md`, `dx.md` | | [`prove-it`](skills/prove-it/SKILL.md) | Reviewing a PR/diff, auditing contract/SLA evidence, or verifying a live status/readiness claim against primary sources | *(Uses Stage 1 acceptance criteria and live system queries)* | | [`swarm-fuse`](skills/swarm-fuse/SKILL.md) | Facilitating multi-agent/group deliberation (`Topic × Participant Context`) when blocked, stalled 10+ minutes, or resolving cross-domain trade-offs | *(Coordinates with `self-challenge` and `deliver-and-verify`)* | | [`unblock`](skills/unblock/SKILL.md) | Restoring progress when work hits a blocker, CI/build failure, runner/env error, `"done but still broken"` report, or upstream dependency defect | *(Coordinates with `swarm-fuse` and `self-challenge`)* | | [`refactor`](skills/refactor/SKILL.md) | Splitting oversized files/modules/test suites, deleting dead code/flags, consolidating duplicate logic into one SSOT, or restructuring into vertical slices | `health.md` | | [`clear-writing`](skills/clear-writing/SKILL.md) | Drafting or rewriting user-facing replies, release notes, technical docs, partner emails, articles, or landing-page copy (`Audience × Purpose × Channel`) | `work-communication.md`, `articles-and-copy.md`, `terminology.md` | | [`author-skill`](skills/author-skill/SKILL.md) | Creating, refactoring, or auditing reusable `SKILL.md` files and their `references/` | `docs.md` | | [`customer-support`](skills/customer-support/SKILL.md) | Replying to incoming customer or user support messages with deterministic state and priority rules | `product-facts.md` | ## Installation Across Agent Harnesses ### 1. Universal Skills CLI (`npx skills` — Claude Code, OpenCode, Codex, Cursor, Windsurf, Antigravity, Amp) ```bash # Install all skills globally across detected coding agents npx skills add SylphxAI/skills -g -y # Or install into the current project workspace npx skills add SylphxAI/skills -y # Update all installed skills to the latest commit on main npx skills update ``` ### 2. Claude Code Native Plugin Marketplace ```bash claude plugin marketplace add SylphxAI/skills claude plugin install sylphx-skills@sylphx ``` To enable automatic background updates in Claude Code (disabled by default for third-party marketplaces): - Run `/plugin` -> open the **Marketplaces** tab -> enable **Auto-update** for `sylphx`, or - Set `"autoUpdate": true` on the `sylphx` entry under `extraKnownMarketplaces` in `.claude/settings.json`. ### 3. OpenCode / Codex / Antigravity Direct Git Clone & Symlink ```bash git clone https://github.com/SylphxAI/skills.git ~/.local/share/sylphx-skills # OpenCode / Codex / Universal (.agents/skills) mkdir -p ~/.agents/skills && ln -sfn ~/.local/share/sylphx-skills/skills/* ~/.agents/skills/ # Claude Code (~/.claude/skills) mkdir -p ~/.claude/skills && ln -sfn ~/.local/share/sylphx-skills/skills/* ~/.claude/skills/ # Antigravity (~/.gemini/antigravity-cli/skills) mkdir -p ~/.gemini/antigravity-cli/skills && ln -sfn ~/.local/share/sylphx-skills/skills/* ~/.gemini/antigravity-cli/skills/ ``` ## Verification Run the repository verification script to validate all YAML frontmatters (`PyYAML`), plugin manifests, relative Markdown links, and reference coverage: ```bash python3 scripts/verify-skills.py python3 scripts/verify-terminology.py python3 -m unittest discover -s scripts -p "test_*.py" ``` ## License [MIT](LICENSE)

Chargement du README du dépôt…

Skills dans ce dépôt

Spécialistes en gestion de projets Développeurs de logiciels Analystes en études de marché et spécialistes en marketing Administrateurs de réseaux et de systèmes informatiques

18 catégories métier · 93% classifié

skill

métier

description

mis à jour

handoff-work

non classé

Safely persist in-progress state and write an actionable resumption brief for another machine, session, or developer. Use when handing off work, switching computers or workspaces, pausing mid-task, saving work-in-progress across environments, or transferring…