Skip to content
Back to search

agent-skills

SkillWorks with CodexGitHub

90 specialist AI agents + 3 project-local extensions for Claude Code / Codex CLI / Antigravity CLI (agy). Anthropic Agent Skills spec-aligned, hub-spoke orchestration via Nexus with 49 Recipes and 11 Skill Packs. Covers development, security, design, testing, FinOps, compliance, observability, and AI/ML.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/simota/agent-skills with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/simota/agent-skills'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/simota--agent-skills/badge.svg)](/listings/simota--agent-skills)

HTML

<a href="/listings/simota--agent-skills"><img src="/api/listings/simota--agent-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Installation

**Recommended — one working tree, symlinked into every CLI**

git clone https://github.com/simota/agent-skills.git ~/repos/agent-skills
cd ~/repos/agent-skills
make link

`make link` symlinks each skill directory into `~/.claude/skills`, `~/.agents/skills`, and `~/.gemini/antigravity-cli/skills` individually, so every CLI reads the same working tree and a single `git pull` updates all of them. Skills already present in those directories are preserved: a name that already exists is reported and skipped, never overwritten. Missing destination parents are skipped except for the guarded Codex migration described below, so the same command works on a single-CLI machine. A directory is an installation hint, not a live CLI availability test.

| Command | Effect | |---------|--------| | `make link` | Link into all three CLI directories — `link-claude` / `link-codex` / `link-agy` for just one | | `make status` | Show the link count per CLI (`linked / linkable`) and how many entries each directory holds | | `make unlink` | Remove only links created by `make link`; other skills and custom aliases stay in place |

**Alternative — clone straight into a CLI directory**

# For Claude Code
git clone https://github.com/simota/agent-skills.git ~/.claude/skills

# For other platforms
git clone https://github.com/simota/agent-skills.git /path/to/your/skills

The shortest path when the target directory does not exist yet and only one CLI needs the skills.

Usage

/Nexus I want to implement a login feature
/Scout Investigate the cause of this bug
/Radar Improve test coverage
/Vision I want to redesign the dashboard with a modern look

Runtime compatibility and legacy Codex installs

Current model/CLI facts and official sources live in [`_common/CLI_COMPATIBILITY.md`](_common/CLI_COMPATIBILITY.md). Skills remain model-agnostic; actual availability, permissions and loading require runtime verification. Gemini CLI and Cursor are secondary readers, not additional installer targets.

Codex skills now use `~/.agents/skills`; settings still use `~/.codex`. The installer preserves old files. To remove **only this repository's old links**, run `make unlink-codex CODEX_DIR="$HOME/.codex/skills"`, inspect the result, then `make link-codex`. Foreign files/links are not removed. If `.agents` is absent but `.codex` exists, the installer creates the new parent only outside the source repository; otherwise create the intended external skill root explicitly.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 686 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # AI Agent Skills
  2. [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
  3. [![Global Agents](https://img.shields.io/badge/Global_Agents-90-blue.svg)]()
  4. [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](CONTRIBUTING.md)
  5. ## Features
  6. - 90 Global Agents + 3 Project-local Extensions: the repository's own operating skills (`orbit`, `lore`, `darwin`) stay out of global profiles
  7. - Nexus Orchestrator: designs the agent chain for a task and runs it
  8. - Platform Agnostic: works with Claude Code, Codex CLI, Antigravity CLI, and others
  9. ## Quick Start
  10. ### Installation
  11. **Recommended — one working tree, symlinked into every CLI**
  12. ```bash
  13. git clone https://github.com/simota/agent-skills.git ~/repos/agent-skills
  14. cd ~/repos/agent-skills
  15. make link
  16. ```
  17. `make link` symlinks each skill directory into `~/.claude/skills`, `~/.agents/skills`, and
  18. `~/.gemini/antigravity-cli/skills` individually, so every CLI reads the same working tree and a single `git pull`
  19. updates all of them. Skills already present in those directories are preserved: a name that already
  20. exists is reported and skipped, never overwritten. Missing destination parents are skipped except for the guarded Codex migration described below, so the same command works on a single-CLI machine. A directory is an installation hint, not a live CLI availability test.
  21. | Command | Effect |
  22. |---------|--------|
  23. | `make link` | Link into all three CLI directories — `link-claude` / `link-codex` / `link-agy` for just one |
  24. | `make status` | Show the link count per CLI (`linked / linkable`) and how many entries each directory holds |
  25. | `make unlink` | Remove only links created by `make link`; other skills and custom aliases stay in place |
  26. **Alternative — clone straight into a CLI directory**
  27. ```bash
  28. # For Claude Code
  29. git clone https://github.com/simota/agent-skills.git ~/.claude/skills
  30. # For other platforms
  31. git clone https://github.com/simota/agent-skills.git /path/to/your/skills
  32. ```
  33. The shortest path when the target directory does not exist yet and only one CLI needs the skills.
  34. ### Usage
  35. ```
  36. /Nexus I want to implement a login feature
  37. /Scout Investigate the cause of this bug
  38. /Radar Improve test coverage
  39. /Vision I want to redesign the dashboard with a modern look
  40. ```
  41. ### Runtime compatibility and legacy Codex installs
  42. Current model/CLI facts and official sources live in [`_common/CLI_COMPATIBILITY.md`](_common/CLI_COMPATIBILITY.md). Skills remain model-agnostic; actual availability, permissions and loading require runtime verification. Gemini CLI and Cursor are secondary readers, not additional installer targets.
  43. Codex skills now use `~/.agents/skills`; settings still use `~/.codex`. The installer preserves old files. To remove **only this repository's old links**, run `make unlink-codex CODEX_DIR="$HOME/.codex/skills"`, inspect the result, then `make link-codex`. Foreign files/links are not removed. If `.agents` is absent but `.codex` exists, the installer creates the new parent only outside the source repository; otherwise create the intended external skill root explicitly.
  44. ## Overview
  45. Each skill specializes in one domain and is coordinated by the Nexus orchestrator when available in the active profile.
  46. ### Roster Counting
  47. - **90 global skills** are the active, reusable roster under `<skill-name>/SKILL.md`.
  48. - **3 project-local extensions** (`orbit`, `lore`, and `darwin`) are repository operating skills and are not included in the global count or global profiles.
  49. - `.claude/skills/` is the canonical project-local location; `.agents/skills/` is a cross-tool mirror, not an additional set of skills.
  50. - `.archive/` holds a retired skill only while `prune`'s sunset protocol keeps it, and is excluded from every active-roster count. The directory does not exist until a sunset creates it; earlier retired packages are recoverable from git history.
  51. The active repository surface is therefore **90 global + 3 project-local = 93 distinct skills**.
  52. ## Agent Catalog
  53. > Category-by-category catalog for 90 global agents plus the 3 project-local extensions marked below.
  54. ### Orchestration
  55. | Agent | Description | Output |
  56. |-------|-------------|--------|
  57. | **Nexus** | _"The right agent at the right time changes everything."_ - Team orchestrator and scope-adaptive product delivery owner. Designs and executes minimum viable agent chains | Prompts, progress management |
  58. | **Sherpa** | _"The mountain doesn't care about your deadline. Plan accordingly."_ - Task decomposition guide. Breaks complex tasks into atomic steps completable within 15 minutes | Checklists |
  59. | **Architect** | _"Every agent is a possibility. Every SKILL.md is a birth certificate."_ - Meta-designer that creates new skill agents. Ecosystem gap analysis, duplication detection, SKILL.md generation | SKILL.md, references |
  60. | **Rally** | _"One task, many hands. Parallel by design."_ - Multi-session parallel orchestrator. Spawns and manages multiple Claude instances via Claude Code Agent Teams API, or Codex CLI subagents, for concurrent task execution | Team management, parallel execution |
  61. | **Sigil** | _"Every project has patterns waiting to become power."_ - Project-local operating-layer designer and generator. Creates skills, recipes, workflows, and a routing map | Project-specific skills |
  62. | **Orbit** *(project-local)* | _"Give me a goal. I'll give you a runner that finishes."_ - Nexus-autoloop completion specialist for this repository. Generates completion scripts, operation contracts, and audits for autonomous loop execution | Runner scripts, contracts |
  63. | **Darwin** *(project-local)* | _"Ecosystems that cannot sense themselves cannot evolve themselves."_ - Repository-local ecosystem evolution orchestrator persisted through `.agents/ECOSYSTEM.md` | Ecosystem Fitness Score, evolution proposals |
  64. | **Lore** *(project-local)* | _"Forgotten lessons are lessons repeated. Institutional memory is the compound interest of experience."_ - Repository-local curator that synthesizes `.agents/*.md` into `METAPATTERNS.md` | METAPATTERNS.md, knowledge insights |
  65. | **Gauge** | _"What gets measured gets managed. What gets audited gets normalized."_ - SKILL.md normalization auditor and self-evolving compliance agent. Scans all skills against the 21-item checklist, classifies violations with P0-P3 priority, generates concrete fix snippets, and evolves detection patterns via web research. No code written | Compliance reports, fix plans, dashboards |
  66. | **Atelier** | _"Design decided upstream. Assets produced downstream. atelier is the studio floor in between."_ - Design-to-implementation pipeline orchestrator for the code-to-visual-to-code closed loop. Coordinates Vision → Muse/Frame → Forge → Artisan → Vitrine → Canvas to deliver design extraction, prototypes, visual assets, slides, and production together while persisting a project design system across downstream agents | Design system package, integrated deliverables |
  67. | **Compass** | _"When in doubt, ask Compass. It finds the right skill for the task."_ - Skill ecosystem navigator and onboarding guide. Lists global agents and available project-local extensions, recommends best fit, and helps newcomers discover the right specialist | Recommendations, agent maps |
  68. | **Prune** | _"A garden grows by what you cut, not what you plant."_ - Ecosystem cleanup auditor. Audits the agent roster for overlap, redundancy, and inactivity, then proposes merge candidates and sunset plans with evidence and archive instructions. Propose-only, no execution | Merge/sunset proposals, archive instructions |
  69. ### Investigation & Planning (Non-coding)
  70. | Agent | Description | Output |
  71. |-------|-------------|--------|
  72. | **Scout** | _"Every bug has a story. I read the ending first."_ - Bug investigation and root cause analysis (RCA). Identifies reproduction steps and fix locations | Investigation report |
  73. | **Ripple** | _"Every change sends ripples. Know where they land before you leap."_ - Pre-change impact analysis. Evaluates risk from both vertical (dependencies, affected files) and horizontal (pattern consistency, naming conventions) perspectives | Impact analysis report |
  74. | **Spark** | _"The best feature is the one users didn't know they needed."_ - Feature proposals. Suggests features built on existing data/logic as Markdown specs | Specification document |
  75. | **Compete** | _"Know your enemy. Know the market. Know yourself."_ - Competitive research, differentiation, positioning, and engineering-professional branding across GitHub, LinkedIn, portfolios, publishing, and conference presence | Competitive analysis, positioning and brand strategy |
  76. | **Voice** | _"Feedback is a gift. Analysis is unwrapping it."_ - User feedback collection, NPS survey design, sentiment analysis, and insight extraction | Feedback report |
  77. | **Field** | _"Good research asks the right questions. Great research changes what you thought was the question."_ - User research design, interview guides, qualitative analysis, persona/journey map creation | Research report |
  78. | **Trace** | _"Every click tells a story. I read between the actions."_ - Session replay analysis, per-persona behavioral pattern extraction, UX problem storytelling. Works with Field/Echo | Behavioral analysis report |
  79. | **Canon** | _"Standards are the accumulated wisdom of the industry. Apply them, don't reinvent them."_ - Standards, regulatory-control, and legal-document review. Evaluates OWASP/WCAG/OpenAPI/ISO 25010, SOC2/PCI-DSS/HIPAA/ISO 27001, ToS, privacy policies, DPA/EULA, cookie consent, and advertising claims | Compliance and legal-review report |
  80. | **Lens** | _"See the code, not just search it."_ - Codebase comprehension specialist. Systematically investigates code structure, feature exploration, and data flow tracing for questions like "Does feature X exist?", "How does flow Y work?", "What is this module's responsibility?" | Investigation report |
  81. | **Magi** | _"Three minds, one verdict. Consensus through diversity."_ - Multi-perspective decisions, founder office hours, bottleneck triage, pitch critique, advisory lenses grounded in documented notable-figure thinking, and business scenario simulation (`simulate` Recipe) | Decision and advisory report |
  82. | **Flux** | _"Bend the light. See what was always there."_ - Thinking refraction engine. Challenges assumptions, combines cross-domain knowledge, and shifts perspectives to produce reframed problem statements. Cynefin-based framework selection, Serendipity Injection, 10+ thinking frameworks, and multi-turn ideation across Expand/Propose/Evaluate/Subtract (`ideate` Recipe). No code written | Reframing package, Insight Matrix, Blind Spot Report |
  83. | **Cast** | _"Personas are not invented. They are discovered, born, and evolved."_ - Persona casting agent. Rapid generation, registry management, lifecycle tracking, and cross-agent distribution of personas from diverse inputs | Persona registry |
  84. | **Matrix** | _"Infinite combinations, finite resources. Matrix finds the minimum that covers the maximum."_ - Universal multi-dimensional analysis agent. Controls combinatorial explosion from arbitrary axis×value inputs. Minimum coverage set selection, execution planning, and prioritization across all domains (testing, deployment, UX validation, risk assessment, compatibility). No code written | Matrix analysis, coverage optimization plan |
  85. | **Saga** | _"Facts are remembered 5-10% of the time. Stories raise that to 65-70%. The customer is the hero. The product is the guide."_ - Narrative design agent. Structures product and feature use cases as customer-centric stories. StoryBrand SB7, Pixar Story Spine, Hero's Journey, JTBD frameworks, pitch narratives, onboarding stories, transformation arcs. No code written | Narrative document |
  86. | **Omen** | _"Plan for the worst. Build for the best."_ - Pre-mortem analysis and failure mode enumeration. Systematically identifies failure scenarios, scores with RPN/AP. No code written | Pre-mortem report |
  87. | **Rank** | _"Every priority tells a story of trade-offs."_ - Priority quantification. Scores and orders competing items using ICE/RICE/WSJF/MoSCoW/Kano frameworks. No code written | Priority report |
  88. | **PDM** | _"Show where the project stands — planned, built, and the gap between."_ - Project delivery status navigator (PdM-style, read-only). Reconciles planned scope (specs/issues/roadmap/PRD) against implemented code to produce feature inventories, unimplemented-feature lists, roadmap rollups, and WBS views | Feature inventory, gap list, roadmap rollup |
  89. **Scout > Ripple > Builder chain**: Scout (bug investigation) > Ripple (fix impact analysis) > Builder (implementation)
  90. **Ripple > Guardian chain**: Ripple (impact analysis) > Guardian (PR strategy)
  91. **Field > Trace > Echo chain**: Field (persona definition) > Trace (real-data validation) > Echo (simulation confirmation)
  92. **Sentinel > Canon > Builder chain**: Sentinel (vulnerability detection) > Canon (OWASP compliance evaluation) > Builder (fix implementation)
  93. **Gateway > Canon > Gateway chain**: Gateway (API design) > Canon (OpenAPI/RFC compliance check) > Gateway (corrections)
  94. **Echo > Canon > Palette chain**: Echo (UX issues) > Canon (WCAG compliance evaluation) > Palette (accessibility fixes)
  95. **Field > Cast > Echo chain**: Field (research data) > Cast (persona integration) > Echo (UI validation)
  96. **Trace > Cast chain**: Trace (behavioral data) > Cast (persona evolution)
  97. ### Git/PR Management
  98. | Agent | Description | Output |
  99. |-------|-------------|--------|
  100. | **Guardian** | _"Every commit tells a story. Make it worth reading."_ - Git/PR gatekeeper. Signal/Noise analysis of changes, commit granularity optimization, branch naming, PR strategy proposals; the `ship` Recipe also creates, CI-watches, and merges the PR (merge is Ask First) | Analysis report, PR preparation |
  101. | **Launch** | _"Shipping is not the end. It's the beginning of accountability."_ - Release management plus read-only GitHub PR collection, weekly/monthly/client reporting, DORA/SPACE metrics, retrospectives, and PDF export | Release plans, CHANGELOG and delivery reports |
  102. | **Trail** | _"Every bug has a birthday. Every regression has a parent commit. Find them."_ - Git history investigation, regression root cause analysis, code archaeology. Travels back in time to uncover the truth | History investigation report |
  103. **Guardian > Judge > Zen chain**: Guardian (PR preparation) > Judge (review) > Zen (fixes)
  104. **Guardian > Launch chain**: Guardian (change analysis) > Launch (release plan)
  105. **Trail > Scout chain**: Trail (regression identification) > Scout (detailed investigation)
  106. ### Quality Assurance
  107. | Agent | Description | Output |
  108. |-------|-------------|--------|
  109. | **Radar** | _"Untested code is unfinished code."_ - Unit/integration test addition, flaky test fixing, coverage improvement, and test-data factories/fixtures (`fixtures` Recipe) | Test code |
  110. | **Voyager** | _"E2E tests are the user's advocate in CI/CD."_ - Cross-platform and iOS E2E specialist. Playwright/Cypress/Appium/Detox/Maestro plus XCUITest and App Store snapshots | E2E test code |
  111. | **Sentinel** | _"Security is not a feature. It's a responsibility."_ - Static security analysis (SAST), vulnerability pattern detection, input validation | Security fixes |
  112. | **Probe** | _"A system is only as secure as its weakest endpoint."_ - Dynamic security testing (DAST), OWASP ZAP/Nuclei integration, penetration testing | Vulnerability report |
  113. | **Vigil** | _"An undetected attack is an undefended system. Vigil ensures nothing passes unseen."_ - Detection Engineering agent. Sigma/YARA rule design, detection coverage mapping (MITRE ATT&CK), threat hunting hypothesis design, Purple Team Blue side execution, Detection-as-Code CI/CD integration | Detection rules, coverage maps |
  114. | **Judge** | _"Good code needs no defense. Bad code has no excuse."_ - Multi-engine code review (Claude + Codex), automated PR review, pre-commit checks, AI hallucination detection | Review report |
  115. | **Zen** | _"Clean code is not written. It's rewritten."_ - Refactoring and code quality improvement (behavior unchanged) | Code improvements |
  116. | **Sweep** | _"Dead code is technical debt that earns no interest."_ - Unused file detection, dead code identification, orphaned file discovery, safe deletion proposals | Cleanup proposals |
  117. | **Attest** | _"Specs are truth. Code is evidence. Attest finds the gaps."_ - Specification compliance verifier. Extracts acceptance criteria from specs, generates BDD scenarios, and adversarially probes for gaps between spec and implementation. Issues CERTIFIED/CONDITIONAL/REJECTED verdicts | Compliance report, BDD scenarios |
  118. | **Siege** | _"Break it before users do. Fix it before they notice."_ - Advanced testing specialist. Load testing (k6/Locust/Artillery), contract testing (Pact CDC), chaos engineering, mutation testing, resilience pattern verification | Test results, resilience reports |
  119. | **Void** | _"The best code is the code that was never written."_ - YAGNI enforcement, scope cutting, complexity reduction proposals. Challenges existence of every feature/abstraction with 5 questions and Cost-of-Keeping Score | Subtraction proposals |
  120. | **Breach** | _"Think like an attacker. Defend like an engineer."_ - Red team engineering. Attack scenario design, threat modeling, MITRE ATT&CK/OWASP frameworks, Purple Team exercises, AI/LLM red teaming | Security assessment |
  121. | **Cloak** | _"Privacy is not a feature. It's a right."_ - Privacy engineering and data governance. PII detection, data flow mapping, consent management, GDPR/CCPA-compliant code implementation | Privacy assessment |
  122. | **Chain** | _"Treat every third-party skill like an npm install. Audit before invoking."_ - Skill/plugin/MCP intake audit plus live npm/PyPI malware investigation. Generates sha256 manifests, scans Unicode and exfiltration patterns, matches campaign IoCs, and produces persistence-first eradication and credential-rotation runbooks | Supply-chain audit, infection report and recovery runbook |
  123. ### Implementation
  124. | Agent | Description | Output |
  125. |-------|-------------|--------|
  126. | **Builder** | _"Types are contracts. Code is a promise."_ - Type-safe production implementation plus Gemini image-generation/editing code, prompt design, batch generation, style transfer, post-processing, provenance, and content-policy gates, plus CLI/TUI construction (`cli`) and regex/parser/DSL authoring (`grammar`) | Production and image-generation code |
  127. | **Artisan** | _"Prototypes promise. Production delivers."_ - Production frontend implementation craftsman. React/Vue/Svelte, Hooks design, state management, Server Components, form handling, data fetching | Frontend code |
  128. | **Forge** | _"Done is better than perfect. Ship it, learn, iterate."_ - Prototyping. Prioritizes working software over perfection. Outputs types.ts, errors.ts, forge-insights.md for Builder handoff | MVP/PoC |
  129. | **Native** | _"Every pixel ships. Every platform matters."_ - Pure-native implementation specialist for iOS (Swift 6.3 + SwiftUI + Liquid Glass), Android (Kotlin 2.4+ + Jetpack Compose + Material 3 Expressive), and macOS. Production-quality features with @Observable / Swift Concurrency, Compose Strong Skipping + Type-safe Navigation, SwiftData / Room, Credential Manager + Passkeys, Privacy Manifest, edge-to-edge, predictive back, Live Activities, App Intents, Foundation Models / Gemini Nano, store compliance, and per-store staged rollout. React Native / Flutter / KMP / CMP are out of scope | Code |
  130. | **Pixel** | _"Every pixel matters. Fidelity is non-negotiable."_ - Faithful reproduction agent. Generates pixel-accurate HTML/CSS from image mockups (PNG/JPG/screenshots) and performs visual verification | HTML/CSS code |
  131. ### AI/ML
  132. | Agent | Description | Output |
  133. |-------|-------------|--------|
  134. | **Oracle** | _"AI is only as good as its architecture. Design it, measure it, trust nothing."_ - AI/ML design and evaluation specialist. Prompt engineering, RAG architecture, LLM application patterns, safety guardrails, evaluation frameworks, MLOps, cost optimization | Design specs, evaluation reports |
  135. | **Chisel** | _"A vague word is a decision you left to chance. Carve it into something that can be checked."_ - Prompt-to-executable-specification translator. Detects vague quality/quantity/style/design/judgment wording, dissolves persona and title lines into capabilities, reconciles contradictory instructions, and records what was deliberately left open | Ambiguity ledger, specified prompt |
  136. **Oracle > Builder > Radar chain**: Oracle (AI/ML design) > Builder (implementation) > Radar (tests)
  137. **Oracle > Stream > Builder chain**: Oracle (RAG design) > Stream (data pipeline) > Builder (implementation)
  138. **Oracle > Sentinel > Oracle chain**: Oracle (safety design) > Sentinel (security review) > Oracle (refinement)
  139. ### Performance
  140. | Agent | Description | Output |
  141. |-------|-------------|--------|
  142. | **Bolt** | _"Speed is a feature. Slowness is a bug you haven't fixed yet."_ - Application performance improvement. Frontend (re-render reduction) and backend (N+1 fix) optimization, plus auto-tuning of GC/threadpool/cache/worker settings | Optimized code |
  143. | **Tuner** | _"A fast query is a happy user. A slow query is a lost customer."_ - DB performance optimization. EXPLAIN ANALYZE analysis, index recommendations, slow query improvement | Query optimization |
  144. ### Observability/SRE
  145. | Agent | Description | Output |
  146. |-------|-------------|--------|
  147. | **Beacon** | _"You can't fix what you can't see. You can't see what you don't measure."_ - Observability and reliability engineering specialist. SLO/SLI design, distributed tracing, alerting strategy, dashboard design, capacity planning, toil automation | SLO definitions, observability specs |
  148. **Beacon > Gear > Builder chain**: Beacon (observability design) > Gear (monitoring implementation) > Builder (instrumentation)
  149. **Triage > Beacon > Gear chain**: Triage (incident postmortem) > Beacon (monitoring improvements) > Gear (implementation)
  150. ### UI/UX
  151. | Agent | Description | Output |
  152. |-------|-------------|--------|
  153. | **Vision** | _"Design is not how it looks. Design is how it feels."_ - Creative direction. Design direction decisions, Design System creation, Muse/Palette/Flow/Forge orchestration | Design strategy |
  154. | **Palette** | _"Usability is invisible when done right, painful when done wrong."_ - Usability improvement, cognitive load reduction, a11y support | UX improvements |
  155. | **Muse** | _"Tokens are the DNA of design. Mutate them with care."_ - Design token application, spacing/border-radius/shadow unification, dark mode support | Visual improvements |
  156. | **Flow** | _"Motion creates emotion. Animation breathes life."_ - UI animation, hover effects, loading states, modal transitions | Animations |
  157. | **Echo** | _"I don't test interfaces. I feel what users feel."_ - Persona validation and synthetic demand. Reports UI confusion, feature requests, JTBD, and unmet needs | UX report |
  158. | **Vitrine** | _"Components without stories are components without context."_ - Storybook story creation, catalog management, Visual Regression integration. CSF 3.0 format | Storybook Stories |
  159. | **Prose** | _"Words are the smallest unit of design. Get them wrong, and nothing else matters."_ - User-facing text specialist. Microcopy, error messages, voice & tone framework, onboarding copy, accessibility text | Copy guidelines, content specs |
  160. | **Frame** | _"Design speaks in pixels. I translate it to code."_ - Figma MCP Server bridge agent. Extracts and structures design context from Figma for implementation agents. Design-to-code bridging, Code Connect management, design system rule extraction. No code written | Structured design context, design system rules |
  161. | **Ink** | _"Every stroke serves a purpose."_ - SVG icon/illustration generation, icon system design, and sprite symbol construction | SVG assets |
  162. ### Documentation
  163. | Agent | Description | Output |
  164. |-------|-------------|--------|
  165. | **Scribe** | _"A specification is a contract between vision and reality."_ - Specification author for PRD/SRS/HLD/LLD and cross-team L0-L4 packages (`unified` Recipe), including implementation checklists, test specifications, and document format conversion (`convert` Recipe) | Specs, design docs |
  166. | **Quill** | _"Code tells computers what to do. Documentation tells humans why."_ - JSDoc/TSDoc additions, README updates, typing `any` to proper type definitions | Documentation |
  167. | **Tome** | _"Changes are forgotten. Knowledge endures."_ - Transforms changes into learning documentation and verified knowledge into public technical articles for note/Zenn/Qiita/dev.to, including article series and repurposing | Learning docs and technical articles |
  168. **Scribe vs Quill vs Tome responsibilities**:
  169. - **Scribe**: Project documentation (PRD, SRS, design docs, checklists, test specifications) and format conversion (Markdown > PDF/Word/HTML, `convert` Recipe)
  170. - **Quill**: Code documentation (JSDoc/TSDoc, README, type definitions)
  171. - **Tome**: Change-based learning materials and external-facing articles for note/Zenn/Qiita/dev.to, including series management
  172. ### Visualization
  173. | Agent | Description | Output |
  174. |-------|-------------|--------|
  175. | **Canvas** | _"A diagram is worth a thousand lines of documentation."_ - Design visualization. Converts code, specs, and context into Mermaid diagrams or ASCII art (flowcharts, sequence diagrams, state machines, class diagrams, ER diagrams, etc.) | Mermaid diagrams / ASCII Art |
  176. | **Cue** | _"Every frame tells a story."_ - Video scripts, storyboards, narration, and Playwright feature-demo production with multi-aspect recording, captions, quality gates, voiceover, transcript, and thumbnails | Video plans, demos and publication assets |
  177. | **Stage** | _"Every slide is a stage."_ - Marp/reveal.js/Slidev slide generation, story composition design, conference talk optimization | Slide decks |
  178. ### Architecture
  179. | Agent | Description | Output |
  180. |-------|-------------|--------|
  181. | **Atlas** | _"Dependencies are destiny. Map them before they map you."_ - Dependency analysis, circular reference detection, ADR/RFC creation | Design documents |
  182. | **Port** | _"From web to native. Translate the experience, not just the code."_ - Web-to-native porting design specialist (2026 spec — Liquid Glass / Material 3 Expressive / Swift 6.3 / targetSdk 36 / Privacy Manifest / 5.1.2(i) AI disclosure aware). Designs porting blueprints from Web (React/Vue/Svelte/Angular) to iOS Swift / Android Kotlin pure-native. Produces feature parity matrices, native architecture maps, regulatory-compliance plans, and Strangler-Fig phased roadmaps. Optionally proposes a hybrid path (Pure-Native UI + KMP shared logic) | Porting blueprint, parity matrix, roadmap |
  183. | **Gateway** | _"APIs are promises to the future. Design them like contracts."_ - API design, review, OpenAPI spec generation, versioning strategy, breaking change detection, and messaging-platform integration (`messaging` Recipe) | API specifications |
  184. | **Grove** | _"A well-structured repository is a well-structured mind."_ - Human- and LLM-optimized repository structure design, including context efficiency, prompt-cache topology, and sharding | Structure design, audit reports |
  185. | **Weave** | _"Every state tells a story. Every transition is a contract."_ - Workflow and state machine design specialist. State transition design, invalid transition detection, Saga patterns, approval flows, and cron/timezone/business-calendar scheduling (`schedule` Recipe) | Designs, diagrams |
  186. | **Seek** | _"The right result at the right time in the right order."_ - Search engine and vector DB design specialist. Full-text search, vector search, hybrid search, RAG retrieval layer | Code, configs |
  187. | **Crypt** | _"Trust no channel. Verify every key."_ - Cryptographic architecture design: algorithm selection, key management, E2E encryption, KMS integration, TLS configuration | Crypto design specs |
  188. **Gateway > Builder > Radar chain**: Gateway (messaging design) > Builder (implementation) > Radar (tests)
  189. ### Data
  190. | Agent | Description | Output |
  191. |-------|-------------|--------|
  192. | **Schema** | _"A schema is a contract with the future."_ - DB schema, migration, ER diagram, and multi-tenant isolation/RLS/routing design | Migrations / schema definitions |
  193. | **Stream** | _"Data flows like water. My job is to build the pipes."_ - Data pipelines. ETL/ELT design, Kafka/Airflow/dbt, batch/streaming selection, data quality management | Pipeline design, DAGs, dbt models |
  194. **Schema > Stream chain**: Schema (data model) > Stream (pipeline design)
  195. ### DevOps
  196. | Agent | Description | Output |
  197. |-------|-------------|--------|
  198. | **Gear** | _"The best CI/CD is the one nobody thinks about."_ - Dependency management, CI/CD and Docker optimization, plus advanced GitHub Actions workflow design | Configuration files |
  199. | **Scaffold** | _"Infrastructure is the silent foundation of every dream."_ - Cloud infrastructure (Terraform/CloudFormation/Pulumi), local dev environments (Docker Compose), IaC design | Infrastructure config |
  200. | **Hone** | _"A sharp blade cuts clean. A sharp config cuts friction."_ - AI CLI configuration auditor and Claude Code lifecycle-hook specialist. Audits Codex/Antigravity/Claude configuration and designs, configures, or debugs scoped hooks with verification. Also owns the personal environment — dotfiles, shell/editor config, and macOS automation (`env` / `automate` Recipes) | Audit reports, proposals and hook configurations |
  201. | **Ledger** | _"Every dollar has a story. Make it a short one."_ - FinOps and cloud cost optimization. IaC cost estimation, right-sizing, RI/SP recommendations, cost anomaly detection | Reports, configs |
  202. | **Shift** | _"Migration is not moving. It's transforming."_ - Migration and upgrade orchestrator. Framework, library, API, database, and infrastructure migrations end-to-end with codemod generation and incremental strategies | Migration plans |
  203. **Builder vs Hone vs Gear vs Scaffold responsibilities**:
  204. - **Builder**: CLI and TUI construction (`cli` Recipe) — the tool itself
  205. - **Hone**: Personal environment (dotfiles, shell, editor, terminal, macOS automation) plus AI CLI configuration audit and Claude Code event-hook design/configuration/debugging
  206. - **Gear**: Project-level DevOps (CI/CD, Docker, monitoring, Git hooks); `gha` mode owns advanced GitHub Actions
  207. - **Scaffold**: Infrastructure provisioning (cloud, Docker Compose, IaC)
  208. ### Internationalization
  209. | Agent | Description | Output |
  210. |-------|-------------|--------|
  211. | **Polyglot** | _"Every language deserves respect. Every user deserves their mother tongue."_ - i18n support. Replaces hardcoded strings with t() functions, formats dates/currencies with Intl API | i18n implementation |
  212. ### Growth
  213. | Agent | Description | Output |
  214. |-------|-------------|--------|
  215. | **Growth** | _"Traffic without conversion is just expensive vanity."_ - SEO (meta/OGP/JSON-LD), SMO (social share display), CRO (CTA improvement), GEO (AI citation optimization), and retention/re-engagement design (`retention` Recipe) | Growth initiatives |
  216. | **Funnel** | _"Above the fold is your one shot. Make every pixel convert."_ - LP (Landing Page) conversion strategist and premium production orchestrator. Framework-based structure design (AIDA/PAS/BAB/4Ps), hero section craft, CTA placement strategy, social proof hierarchy, mobile-first responsive implementation | LP structure, copy, specs |
  217. ### Analytics
  218. | Agent | Description | Output |
  219. |-------|-------------|--------|
  220. | **Pulse** | _"What gets measured gets managed. What gets measured wrong gets destroyed."_ - KPI definition, tracking event design, dashboard spec creation | Metrics design |
  221. | **Experiment** | _"Every hypothesis deserves a fair trial. Every decision deserves data."_ - A/B test design, hypothesis documentation, sample size calculation, feature flag implementation | Experiment reports |
  222. ### Operations
  223. | Agent | Description | Output |
  224. |-------|-------------|--------|
  225. | **Triage** | _"In chaos, clarity is the first act of healing."_ - Incident response, impact assessment, recovery procedures, postmortem creation | Operations reports |
  226. | **Mend** | _"Known failures deserve known fixes. Speed of recovery defines reliability."_ - Auto-repair agent for known failure patterns. Receives Triage diagnostics and Beacon alerts, executes runbooks based on safety tier classification, performs staged verification, and manages rollback | Repair results, verification reports |
  227. **Triage > Mend > Beacon chain**: Triage (diagnosis) > Mend (auto-repair) > Beacon (monitoring recovery)
  228. ### Browser Automation
  229. | Agent | Description | Output |
  230. |-------|-------------|--------|
  231. | **Vector** | _"The browser is a stage. Every click is a scene."_ - Browser automation via Playwright/Chrome DevTools. Data collection, form interaction, screenshots, network monitoring, and crawl-system architecture (`crawl` Recipe) | Automation scripts |
  232. Cue vs Vitrine responsibilities:
  233. - Cue: Browser (Web UI) demo videos (Playwright, .webm output)
  234. - Vitrine: Component catalogs and Storybook stories with visual-regression wiring
  235. ## Workflows
  236. ### Basic Usage
  237. Invoke an agent with `/AgentName`. It runs the task and suggests a handoff to other agents when one is needed.
  238. ### Orchestration with Nexus
  239. Use Nexus for complex tasks. Nexus operates in the following modes:
  240. | Mode | Trigger | Behavior | Interaction |
  241. |------|---------|----------|-------------|
  242. | **Full Auto** (default) | no marker, or `## NEXUS_AUTORUN_FULL` | Executes all tasks, including COMPLEX, with guardrails | Ask First triggers and guardrail escalations only |
  243. | **Auto** | `## NEXUS_AUTORUN` | Executes simple tasks only; complex tasks fall back to Guided | Ask First triggers, guardrail escalations, and errors |
  244. | **Guided** | `## NEXUS_GUIDED` | Confirms at decision points | Option-based interaction |
  245. | **Interactive** | `## NEXUS_INTERACTIVE` | Confirms at each step | Always interactive |
  246. | **Continue** | `## NEXUS_HANDOFF` | Result handoff | Interaction as needed |
  247. ### Interactive Execution (Guided/Interactive)
  248. Each agent asks for user confirmation at important decision points (using the platform's interaction capabilities):
  249. - Start confirmation (Interactive only): confirms the approach after chain design, before execution; Guided confirms only at decision points
  250. - Decision point confirmation: security risks, destructive changes, multiple approaches, etc.
  251. - Questions as options: choose from 2-4 options ("Other" is always available)
  252. ```yaml
  253. # Example option-based question
  254. questions:
  255. - question: "A potential security vulnerability was found. How would you like to proceed?"
  256. header: "Security"
  257. options:
  258. - label: "Audit with Sentinel (Recommended)"
  259. description: "Request review from the security specialist agent"
  260. - label: "Continue with acknowledged risk"
  261. description: "Proceed at your own risk"
  262. - label: "Abort investigation"
  263. description: "Stop for safety"
  264. ```
  265. ### Automatic Mode Selection by Complexity
  266. | Indicator | SIMPLE | COMPLEX |
  267. |-----------|--------|---------|
  268. | Estimated steps | 1-2 | 6+ (3-5 = MEDIUM) |
  269. | Affected files | 1-3 | 4+ |
  270. | Security-related | No | Yes |
  271. | Destructive changes | No | Yes |
  272. See `_common/INTERACTION.md` for details.
  273. ### Chain Templates by Task Type
  274. > **Source of truth:** base chains (rows without a `/variant`) come from `nexus/reference/routing-matrix.md`; `TYPE/variant` rows come from `nexus/reference/agent-chains.md`. This is a representative excerpt — where it differs, those files win. Variant rows omit the terminal Guardian (SHIP) step that the matrix default adds for code changes; `?` marks an optional step.
  275. #### Investigation & Incidents
  276. | Task | Use | Chain |
  277. |------|-----|-------|
  278. | INVESTIGATE | Does X exist / how does it work | Lens |
  279. | INVESTIGATE/flow | Data/processing flow tracing | Lens > Canvas |
  280. | INVESTIGATE/onboarding | Whole-codebase understanding | Lens > Scribe |
  281. | INVESTIGATE/pre-impl | Investigate, then implement | Lens > Builder > Radar |
  282. | INVESTIGATE/regression | Regression from a past commit | Trail > Scout > Builder > Radar |
  283. | BUG/simple | Simple bug fix | Scout > Lens > Radar > Builder > Radar |
  284. | BUG/complex | Complex bug (RCA required) | Scout > Lens > Sherpa > Radar > Builder > Radar > Sentinel |
  285. | INCIDENT/SEV1/2 | Major incident | Triage > Scout > Builder > Radar > Triage (postmortem) |
  286. > **Lens vs Scout**: Lens = codebase understanding and feature exploration ("Does X exist?", "How does Y flow?"), Scout = bug investigation and root cause analysis ("Why did it break?")
  287. #### Features, APIs & Integrations
  288. | Task | Use | Chain |
  289. |------|-----|-------|
  290. | FEATURE/S | Small feature | Builder > Radar |
  291. | FEATURE/M | Medium feature | Sherpa > Forge > Builder > Radar |
  292. | FEATURE/L | Large feature | Spark > Sherpa > Forge > Builder > Radar > Quill |
  293. | FEATURE/UI | UI-heavy feature | Spark > Forge > Muse > Builder > Lens > Radar |
  294. | FEATURE/frontend | Frontend feature | Forge > Artisan > Radar |
  295. | API/new | New API | Gateway > Builder > Radar > Quill |
  296. | MESSAGING | Chat bots, channel adapters, WebSocket/SSE | Gateway > Builder > Radar |
  297. | WEBHOOK | Webhook contracts and handlers | Gateway > Builder |
  298. #### Design & UX
  299. | Task | Use | Chain |
  300. |------|-----|-------|
  301. | DESIGN/creative | New visual direction to UI | Vision > Muse > Forge > Artisan > Radar |
  302. | DESIGN/flow | Animation / interaction | Flow > Artisan > Radar |
  303. | DESIGN/usability-audit | Usability fixes | Palette > Artisan > Radar |
  304. | DESIGN_SYSTEM/full | Design system build-out | Vision > Muse > Vitrine > Artisan > Quill > Radar |
  305. | UX_RESEARCH | UX research | Field > Echo > Palette |
  306. | UX_RESEARCH/session-replay | Session analysis | Trace > Field > Echo > Palette |
  307. #### Refactoring, Performance & Quality
  308. | Task | Use | Chain |
  309. |------|-----|-------|
  310. | REFACTOR/small | Small refactor | Radar? > Zen > Radar |
  311. | REFACTOR/arch | Architecture improvement | Atlas > Sherpa > Zen > Radar |
  312. | OPTIMIZE/app | Application hotspot | Bolt > Radar |
  313. | OPTIMIZE/db | Database optimization | Tuner > Schema > Builder > Radar |
  314. | YAGNI/standard | Subtract over-engineering | Void > Sweep > Zen > Radar |
  315. | ANALYSIS | Change impact analysis | Ripple > Builder > Radar |
  316. | ANALYSIS/cleanup | Code cleanup | Sweep > Zen > Radar |
  317. | QUALITY/standard | Quality audit | Judge > Zen > Radar > Sentinel > Canvas |
  318. #### Security, Testing & Review
  319. | Task | Use | Chain |
  320. |------|-----|-------|
  321. | SECURITY/static | Static analysis | Sentinel > Builder > Radar > Sentinel |
  322. | SECURITY/dynamic | Dynamic testing | Sentinel > Probe > Builder > Radar > Probe |
  323. | SECURITY/full | Full audit | Sentinel > Probe > Builder > Radar > Sentinel > Probe |
  324. | TEST | Unit / coverage | Radar |
  325. | TEST/e2e | E2E tests | Voyager > Lens > Radar |
  326. | LOAD_TEST/chaos | Chaos engineering | Siege > Bolt > Triage > Builder > Beacon |
  327. | REVIEW/PR | PR review | Judge > Builder/Zen/Sentinel (based on findings) > Radar |
  328. #### AI, Operations & Release
  329. | Task | Use | Chain |
  330. |------|-----|-------|
  331. | AI_FEATURE | LLM feature | Oracle > Builder > Radar |
  332. | AI_FEATURE/rag | RAG pipeline | Oracle > Gateway > Builder > Radar |
  333. | PROMPT_SPEC | Vague prompt to executable spec | Chisel |
  334. | OBSERVABILITY/slo-design | SLO definition & monitoring | Beacon > Gear > Builder > Radar |
  335. | INFRA | Infrastructure setup | Scaffold > Gear > Radar |
  336. | GHA_WORKFLOW/new | New GitHub Actions workflow | Gear[gha] > Radar |
  337. | PRERELEASE/full | Pre-release gate | Sentinel > Radar > Guardian > Launch |
  338. | DEPLOY/full | Full release pipeline | Radar > Guardian > Launch |
  339. | MODERNIZE/stack | Tech stack refresh | Lens > Shift (detect+modernize) > Sherpa > Builder > Radar |
  340. #### Decisions, Strategy & Documents
  341. | Task | Use | Chain |
  342. |------|-----|-------|
  343. | DECISION/architecture | Architecture selection | Magi > Builder/Zen (based on verdict) |
  344. | DECISION/strategy | Strategic decision | Scribe[unified] > Magi > Spark |
  345. | ADVISORY | Office-hours advisory | Magi > (Builder/Echo/Sherpa) |
  346. | STRATEGY/compete | Competitive analysis to implementation | Compete > Spark > Builder > Radar |
  347. | MARKETING/quick | Quick marketing plan | Compete > Growth > Funnel |
  348. | REQUIREMENTS/complex | PRD / requirements | Scribe[unified] > Magi > Scribe > Sherpa > Canvas |
  349. | DOCS | Code documentation | Quill |
  350. | DOCS/report | PR report | Launch[weekly] > Scribe |
  351. | PROJECT/full | Full product from an ambiguous goal | Nexus[deliver] > Sherpa > Builder > Radar > Guardian > Launch |
  352. #### Parallel Execution (Rally Integration)
  353. For large-scale tasks where parallel execution is beneficial, Nexus escalates to Rally (`agent-chains.md` § Rally Parallel Chain Variants).
  354. | Base Chain | Parallel Chain |
  355. |------------|----------------|
  356. | FEATURE/L | Spark > Sherpa > Rally(Forge+Artisan, Builder, Radar) |
  357. | FEATURE/M (multi-unit) | Sherpa > Rally(Builder x N, Radar) |
  358. | FEATURE/fullstack | Rally(Artisan, Builder, Radar) |
  359. | BUG/multiple | Rally(Builder x N) > Radar |
  360. | REFACTOR/arch (multi-module) | Atlas > Sherpa > Rally(Zen x N) > Radar |
  361. | SECURITY/full | Rally(Sentinel, Probe) > Builder > Radar |
  362. | TEST/coverage | Rally(Radar, Voyager) |
  363. | MODERNIZE/stack | Lens > Shift (detect+modernize) > Sherpa > Rally(Builder x N) > Radar |
  364. | DOCS/full | Rally(Quill, Canvas, Vitrine) |
  365. > **Rally Escalation Criteria**: Rally is triggered when there are 2+ independent implementation steps, changes span 4+ files across 2+ domains, or Sherpa detects a `parallel_group`.
  366. > **Nexus parallel vs Rally**: Nexus's built-in `_PARALLEL_BRANCHES` is for lightweight parallelism (each branch < 50 lines). Rally's multi-session parallelism is used for substantial implementation work.
  367. > **Nexus build modes**: `feature` handles one bounded capability, `deliver` sizes a product/MVP chain to scope, and `apex` runs a high-investment discovery-to-ship workflow.
  368. ## Shared Knowledge
  369. Agents share knowledge through the `.agents/` directory:
  370. | File | Purpose | When to Update |
  371. |------|---------|---------------|
  372. | `PROJECT.md` | Shared knowledge + activity log | When the task produced a reusable insight or a decision later agents need (`_common/OPERATIONAL.md` § Activity Log) |
  373. | `{agent}.md` | Agent-specific learnings | When domain-specific discoveries are made |
  374. ### PROJECT.md Structure
  375. - Architecture Decisions
  376. - Domain Glossary
  377. - API & External Services: external service constraints
  378. - Known Gotchas
  379. - Browser / Environment Support
  380. - Security Considerations: security constraints
  381. - Performance Budgets: performance targets
  382. - Activity Log: agent work history
  383. - Cross-Agent Discoveries
  384. ## Agent Principles
  385. Shared rules live in `_common/` (`BOUNDARIES.md`, `OPERATIONAL.md`, `INTERACTION.md`, ...); each skill's `## Boundaries` section states its own Always / Ask first / Never rules.
  386. ## Directory Structure
  387. ```
  388. agent-skills/
  389. ├── <skill>/ # one directory per global skill (see Agent Catalog)
  390. │ ├── SKILL.md # entry point: frontmatter, boundaries, recipes
  391. │ └── reference/ # optional on-demand references
  392. ├── .claude/skills/ # project-local skills: darwin, lore, orbit (canonical)
  393. ├── .agents/skills/ # synchronized mirror of .claude/skills/
  394. ├── _common/ # shared protocols (BOUNDARIES.md, HANDOFF.md, ...)
  395. │ └── scripts/ # checkers behind `make validate` / `make test`
  396. ├── _templates/ # SKILL_TEMPLATE.md, PROJECT.md, kits
  397. └── Makefile # link / unlink / status / validate / test / check / hooks
  398. ```
  399. ## Usage Examples
  400. ### Multi-Agent Collaboration (Nexus)
  401. #### New Feature Development (Auto-execution)
  402. ```
  403. /Nexus
  404. I want to add a user profile editing feature
  405. - Edit name, email, and avatar image
  406. - With validation
  407. - Show toast on successful save
  408. ## NEXUS_AUTORUN_FULL
  409. ```
  410. **Execution chain**: Spark (spec) > Sherpa (task decomposition) > Forge (prototype) > Builder (production implementation) > Radar (tests) > Quill (documentation)
  411. #### Bug Fix (Complex Case)
  412. ```
  413. /Nexus
  414. Investigate and fix a payment error that only occurs in production.
  415. It doesn't reproduce locally.
  416. ## NEXUS_AUTORUN_FULL
  417. ```
  418. **Execution chain**: Scout (investigation) > Sherpa (task decomposition) > Builder (fix) > Radar (regression tests) > Sentinel (security verification)
  419. #### Large-scale Refactoring
  420. ```
  421. /Nexus
  422. The authentication module has become spaghetti code.
  423. I want to refactor it to follow Clean Architecture.
  424. ```
  425. **Execution chain**: Atlas (architecture design) > Sherpa (phased plan) > Zen (refactoring) > Radar (tests)
  426. #### UI Feature Addition
  427. ```
  428. /Nexus
  429. Add a review submission feature to the product detail page.
  430. I want users to be able to input star ratings and comments.
  431. ```
  432. **Execution chain**: Spark (spec) > Forge (UI prototype) > Muse (design adjustment) > Builder (implementation) > Radar (tests)
  433. ### Step-by-Step Execution (Manual Control)
  434. #### Gradual Feature Development
  435. ```
  436. /Nexus
  437. I want to add a search feature.
  438. ## NEXUS_INTERACTIVE
  439. ```
  440. `## NEXUS_INTERACTIVE` confirms before every step; `/Nexus` with no marker runs as AUTORUN_FULL.
  441. #### AUTORUN Mode (Automatic Execution - Recommended)
  442. ```
  443. ## NEXUS_AUTORUN_FULL
  444. I want to add a search feature.
  445. ```
  446. Nexus automatically executes the chain:
  447. ```
  448. ## Nexus Execution: Add search feature
  449. - Chain: **Spark** > **Sherpa** > **Builder** > **Radar**
  450. - Mode: AUTORUN_FULL
  451. ### Executing Step 1/4: Spark
  452. _AGENT_CONTEXT:
  453. Role: Spark
  454. Task: Create search feature specification
  455. [Creating specification as Spark...]
  456. _STEP_COMPLETE:
  457. Agent: Spark
  458. Status: SUCCESS
  459. Next: Sherpa
  460. ### Executing Step 2/4: Sherpa
  461. [Automatically proceeding to next step...]
  462. ```
  463. #### GUIDED Mode (Confirm at Decision Points)
  464. Nexus runs the chain itself and pauses only at decision points (Ask First triggers, guardrail escalations, or a skill's interaction triggers).
  465. ```
  466. ## NEXUS_GUIDED
  467. I want to add a search feature.
  468. ```
  469. Nexus designs the chain, runs it, and stops when a decision needs you:
  470. ```
  471. ## Nexus Plan: Add search feature
  472. - Chain: **Spark** > **Sherpa** > **Builder** > **Radar**
  473. - Current step: 1/4 (Spark)
  474. ### Decision needed
  475. Spark proposed two search scopes. Which should Sherpa plan for?
  476. [Options]
  477. ```
  478. ### Task Decomposition in Practice (Sherpa)
  479. #### Breaking Large Tasks into Small Steps
  480. ```
  481. /Sherpa
  482. I want to implement an e-commerce checkout flow.
  483. ```
  484. **Output**:
  485. ```
  486. ## Sherpa's Guide
  487. **Current Objective:** Checkout flow implementation
  488. **Progress:** 0/8 steps completed
  489. ### NOW: Create type definitions for cart contents
  490. Define CartItem and CartSummary interfaces.
  491. *(Shall I invoke Builder?)*
  492. ### Upcoming Path:
  493. - [ ] Create cart API endpoint
  494. - [ ] Shipping address input form
  495. - [ ] Payment method selection UI
  496. - [ ] Order confirmation screen
  497. - [ ] Order completion processing
  498. - [ ] Error handling
  499. - [ ] Add tests
  500. **Status:** On Track
  501. ```
  502. ## Customization
  503. ### Adding New Agents
  504. Follow [CONTRIBUTING.md § 新しいグローバルエージェントの追加](CONTRIBUTING.md#新しいグローバルエージェントの追加) (adding a new global agent): it lists the required sections, the Complexity Budget declaration, every roster that must be updated by hand, and the generators to re-run. A new skill is done when `make check` passes.
  505. ### Modifying Agents
  506. Edit each `SKILL.md` directly, following [_templates/SKILL_TEMPLATE.md](_templates/SKILL_TEMPLATE.md). Frontmatter example:
  507. ```markdown
  508. ---
  509. name: skill-name
  510. description: Agent description
  511. ---
  512. [Detailed agent instructions]
  513. ```
  514. ## License
  515. MIT

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.