civiltekk-opencode-claude-skills Catalog

Solution-architect review — system design, layer boundaries, dependency direction; transitive blast-radius and consumer impact across modules; owns PLAN atomicity approval. Triggers: architecture review, system design review, blast radius, impact analysis, clean architecture.

Autonomous code optimization loop — modify, verify by benchmark, keep/revert against any metric (coverage, bundle size, runtime, errors). Full repo edit.

Autonomous ML training research loop — modify train.py, train, parse val_bpb, keep/revert. NVIDIA GPU required; path-restricted edits (train.py + research files).

Autonomous literature-review loop (Tier 2, web-only) — fetches papers, structured summaries, living research.md with category tracking. No Bash; research-files-only edits.

CAD/engineering/robotics specialist driving the cad-specialist-skill router — parametric CAD (STEP/STL/3MF/GLB), DXF, drawing redraw (DWG/DXF/PDF/image), URDF/SRDF/SDF, G-code slicing, 3D printing, SendCutSend, Civil 3D, APS, Open3D. Routes internally or installs capability siblings on demand.

Tech-lead pre-commit code review — SOLID, clean code, code smells, object design; severity-gated (BLOCK/WARN/NOTE) with direct-caller verification at diff scope. Triggers: code review, review my code, pre-commit review, quality gate.

Test coverage reporting — badge generation, README updates, threshold enforcement for Next.js and Python.

Runs live client discovery sessions — on-the-fly wireframes, verbatim feedback capture, synthesizes the Vision Document. Triggers: create vision, vision document, concept brief, discovery session, solution vision.

Documentation generation — docstrings, README coverage badges, technical docs per language standards (PEP 257, Javadoc, JSDoc, XML).

Word document creation — create, read, edit, convert .docx with professional formatting, tracked changes, comments, images.

Diagnoses and resolves errors, exceptions, stack traces. Native multimodal (zai-coding-plan/glm-5.3-flash) — reads error screenshots directly. Explicit user invocation only.

Fast codebase exploration — find files by pattern, search code for keywords, answer structural questions.

Image analysis utility — perceives images natively when the runtime provides vision, with a direct Z.AI vision API fallback. Takes paths/URLs; returns bounded structured analysis.

Multi-language code review — Python, TypeScript/JavaScript, Go, Rust, Java: idioms, type safety, concurrency, error handling, security, and framework-specific checks (FastAPI, React/Next.js, Spring Boot, Tokio).

Lint executor — runs linters and auto-fixes detected via language-linting-skill (the rules reference); gate sequence and pass semantics per verification-loop-skill. Python Ruff, JS/TS ESLint, Java Checkstyle/SpotBugs, C# dotnet format/analyzers, generic workflows.

Autonomous loop operator — iterates tasks until completion criteria are met, with self-correction and progress tracking.

Next.js 16 specialist, three modes — project scaffolding (shadcn, Tailwind v4, React Compiler), runtime diagnosis via next-devtools-mcp, proactive audit. Routes to skills by task type.

Routes office document operations (docx, pptx, xlsx) to specialized subagents by file type.

Creates and maintains OpenCode rules (AGENTS.md), agents, skills; scaffolds configurator repos; verifies compliance with opencode.ai docs.

Executes OpenCode v1→v2 migrations — config field mapping, agent/command frontmatter conversion, plugin API triage and ports, MCP/cli.json migration, post-migration verification. Triggers: opencode v2 migration, migrate v1 config, port opencode plugin to v2.

OpenTofu/Terraform infrastructure — Kubernetes, Neon, AWS, Keycloak, ECR provisioning; provider setup with proper state management.

PowerPoint pipeline (generate-template, generate-slide, template-modifier) — fills user-supplied Slide Master templates, embeds hidden JSON schema, image-analyzer verification. Never builds from scratch.

PR workflows with framework-specific quality gates — PR creation, lint/build/test, semantic versioning, JIRA integration.

Git repository operations — setup, release workflows, branch protection, GitHub Actions, labels, semver, PR workflows, git/gh best practices. Triggers: repo setup, branch protection, release workflow, version bump, repo audit, gh labels.

Drafts BRDs (BABOK/IIBA, sponsor scope) and SRS (IEEE 830, internal scope) from discovery interviews; grills requirements gaps surfaced by reviewers (Mode R). Triggers: create brd/srs/prd, business or software requirements, functional spec, specification, requirements gaps from review.

Audits and fixes responsive defects in Next.js pages via Playwright (mobile/tablet/desktop) — tiered fixes (auto-fix, propose+verify, report), re-verification, screenshot review via image-analyzer.

Startup-style PowerPoint presentations — pitch decks, investor slides, board updates.

Startup founder operations — reports, quotations, spreadsheets, presentations, day-to-day business.

Test Driven Development guidance — red-green-refactor cycle across languages and frameworks.

Authors Technical Design Documents — converts SRS/feature specs into architecture, data models, API contracts, ADRs with CodeGraph impact analysis. Triggers: technical design, architecture/system design, design spec.

Test generation — Python pytest, Next.js unit tests, generic frameworks, following best practices.

Review-only UI/UX design review — 13-axis rubric (AslanMazhidov, RNT56, Nielsen, anti-default AI detection) over screenshots, source, live URLs; native multimodal screenshot reading, image-analyzer delegation on failure or request.

Spreadsheet operations (.xlsx, .csv) — create, read, edit, convert tabular data files.

Media generation delegate — images, video, audio transcription, OCR via Z.AI PAYG skills. Runs skills in isolation, saves artifacts to disk, returns file paths. Triggers: generate image/video, transcribe, OCR, extract text.

Implement and audit web accessibility — WCAG 2.1, ARIA, keyboard navigation, contrast, screen readers, axe-core/Lighthouse audits, semantic HTML, focus management.

Debug why agents or skills aren't working as expected with systematic diagnosis, configuration validation, and fix recommendations

Deploy and troubleshoot Next.js 16+ on AWS Amplify Hosting — amplify.yml, SSR Lambda env vars, CloudFront OAC, Route53 DNS, deploy triggers, rollback.

Solution-architect review methodology — decision-tree target routing, layer/dependency review axes, transitive blast-radius gate, finding schema, and severity rubric for architecture and PLAN design review. Loaded by architecture-review-subagent and inline pipeline arms. Triggers: architecture review, blast radius, plan review, inline arch review.

Implement authentication and authorization — OAuth2/OIDC, JWT, sessions, RBAC/ABAC, NextAuth/Auth.js, Passport.js, password hashing, CSRF.

Autonomous code optimization loop — modify, verify by benchmark, keep/revert against any metric (coverage, bundle size, runtime, error count). TDD-flavored.

Canonical autoresearch iteration protocol — Understand, Hypothesize, Experiment, Evaluate, Log loop; mechanical pass/score evaluator; git-as-memory keep/revert; stuck detection.

Autonomous ML training research loop (NVIDIA GPU required) — modify train.py, run fixed-budget experiment, parse val_bpb, keep/revert. karpathy-style.

Autonomous literature-review loop (Tier 2, web-only, no Bash) — fetches papers, extracts structured summaries, builds living research.md with category tracking.

AWS IaC safety patterns — resource count gating, SSM CI passthrough, ECR lifecycle ownership, SSM_SECRETS_REFERENCE, Lambda Web Adapter dual-target.

Find what a change could break beyond the diff and prove the one safety fact by running real code — 5-step evidence ladder. Triggers: blast radius, what could this break, pre-ship safety check.

Route browser work across agent toolsets — probe a managed desktop-browser toolset read-only before any mutating call, fall back to independent automation stacks on disconnect, verify-then-handoff for show-the-user intents. Harness values in references/. Triggers: browser.disconnected, no desktop browser connected, browser toolset, probe before tabs.open, desktop browser attach, browser fallback.

Generate and validate 2D DXF drawings from Python ezdxf sources — profiles, outlines, gaskets, panels, flat patterns, laser/plasma/waterjet cut layouts.

Create, modify, inspect, validate STEP-first parametric CAD parts and assemblies from specs, reference images, or 2D drawings; STEP/STL/3MF/GLB outputs.

Create and render browser-native implicit CAD (.implicit.js/.mjs) — GLSL signed-distance fields, smooth booleans, TPMS fields. Experimental.

Rebuild and validate 2D CAD drawings on Linux — exact redraw from a source DXF/DWG, PDF-to-DXF conversion, or image-to-CAD reconstruction with evidence-aware specs. Triggers: dwg redraw, redraw dxf, image to dxf, pdf to dxf, drawing fingerprint, raster to cad, redraw validation.

CAD, robotics, and hardware router — one entry point for parametric CAD, STEP/STL/3MF/GLB parts, DXF drawings, drawing redraw, URDF/SRDF/SDF robot descriptions (Gazebo, MoveIt), G-code slicing, 3D printing, SendCutSend preflight, Civil 3D, Autodesk APS, Open3D point clouds, and part catalogs. Routes each task to the right specialist route.

Start or reuse CAD Viewer and return review links for CAD, implicit, robot-description, and G-code files (.step, .stl, .glb, .3mf, .gcode, .dxf, .urdf, .srdf, .sdf).

Generate automated changelogs for Python projects using git-cliff with PEP 440 versioning support

Design and document APIs — REST conventions, OpenAPI/Swagger generation, GraphQL schemas, versioning, pagination, rate limiting, error formats, HATEOAS — and detect OpenAPI contract changes, classify breaking vs additive, map consumer impact, generate migration plans (oasdiff). Triggers: openapi diff, api contract, breaking change, contract review, spec changed, regenerate client.

Detect a repo's coding harness(es) — OpenCode v1/v2, pi, Claude Code, Codex — and provision parity project-level setup: shared .agents/skills dir, canonical AGENTS.md with per-harness shims, harness config files, MCP per supported harness. Triggers: set up harness for this repo, harness parity, project harness setup, setup pi/opencode/claude/codex.

Token-overhead audits across agents, skills, and MCP servers (classification + optimization recommendations) and context-compaction strategy that preserves critical information while reducing token usage. Triggers: audit context budget, context budget, token overhead, how much context am I using, audit config bloat, check skill overhead, compact context, summarize session, reduce context, what can we drop, session getting long, preserve key decisions.

Create diagrams — ASCII diagrams from workflow definitions saved as image files (PNG, SVG, etc.), and Mermaid diagrams as Markdown fenced blocks (native GitHub/GitLab/VS Code rendering); optional mmdc CLI for standalone SVG/PNG.

In-session documentation delegate — docstrings and docs for new/changed symbols only, generating language-specific docstrings for C#, Java, Python, and TypeScript (PEP 257, Javadoc, JSDoc, XML). Triggers: inline docs, docstrings in-session, plan-execution-inline-skill documentation step, generate docstrings, convert docstring styles, doc-coverage CI.

Documentation sync and drift auditing for the configurator repo — update setup.sh, setup.ps1, README.md, and AGENTS.md counts and listings when adding or removing skills/subagents; audit and auto-fix doc drift across PLAN, README, AGENTS.md, deploy scripts (count sync, PLAN-vs-reality drift, orphan references). Triggers: sync docs, adding a new skill or subagent, update skill counts, verify documentation accuracy, doc drift, audit documentation consistency, count mismatch, orphan references, stale PLAN checkboxes.

Conventional Commits and compact-commit budgets — type, scope, breaking changes, semver guidance, atomic-commit granularity by layer; 72-char subject, 150-word body, semantic grouping. Triggers: commit and push, make a commit, write a commit message, commit message format, semantic commit, compact commit, concise commit, terse commit. Not for release pipelines (semantic-release-convention-skill).

Catalog install assistant for civiltekk skills, subagents, and dep-shipped plugins — find by intent, install with mandatory dry-run and consent-gated global scope, maintain via update/prune/remove; offers additive AGENTS.md enforcement. Triggers: find a skill, install skill into project, add subagent, update installed skills, is there a skill for, @civiltekk-install-assistant, /civiltekk-install-assistant.

Next.js engineering, four routes. scaffold: scaffold standardized Next.js 16 demos — shadcn, Tailwind v4, src directory with path aliases, React Compiler, Tekk-prefixed components. runtime-diagnosis: next-devtools-mcp — Next.js 16+ runtime diagnosis via MCP (get_errors, get_logs, get_routes, server actions); config and workflows. image-usage: proper Next.js 16 Image component usage with configuration for remote domains, responsive images, and breaking changes from previous versions. threejs: Three.js + Next.js (App Router, React 19) integration — SSR pitfalls, GLSL bundling, hydration, WebGL context loss, R3F/drei, WebXR, companion-library decision tree; version detection first.

Generate OpenCode agents and skills following official documentation best practices — frontmatter contract, lean content standard, permissions, placement, validation. Triggers: create/generate/edit an agent or skill, new subagent, new skill, author agent, author skill. Not for installer tier/model resolution.

OpenTofu/Terraform infrastructure as code, four routes. first-time-setup: configure cloud providers, authentication, and remote state backends (S3, Azure Storage, GCS). plan-apply: provisioning workflow — plan→apply discipline, resource lifecycle, state management and drift. explore: AWS, Kubernetes, Neon Postgres, and Keycloak resources as code. shared-infra-split: provision-once shared modules (ECR/Amplify pattern) vs per-env stacks, and live migrations between the two. Triggers: opentofu, tofu, terraform plan, terraform apply, infrastructure as code, IaC, provider setup, provider pin, state backend, remote state, drift, tofu import, Neon branching, Keycloak realm, Helm release, Kubernetes as code, shared module, provision once, module split, shared infra.

Over-engineering audits and debt tracking — whole-repo audit (ranked delete/stdlib/native/yagni/shrink findings, net removable count, one-shot report), diff-only review (concrete replacements), and `ponytail:` comment harvest into a debt ledger (ceiling + upgrade trigger per marker, no-trigger rot). Triggers: audit for over-engineering, find bloat, what can I delete, over-engineered diff, simplify this change, review for bloat, ponytail shortcuts, what did ponytail defer, tech debt markers.

PR lifecycle workflow, two routes. create: pre-merge pipeline — target branch, framework/language detection, quality checks per the gate memo, structured PR body, semver labels, JIRA image attachments and tracking. merge: post-merge pipeline — merges PR, monitors CI, auto-fixes failures, updates JIRA, deletes source branch; promotions between long-lived lanes run a divergence pre-flight (backmerge PR first, then the promote PR). Triggers: create pr, make pr, open pr, submit pr, ready for pr, create pull request, pr to [branch], 'pr merge to [branch]', 'merge the PR', 'complete the PR', 'promote <branch> to <branch>', 'promote to uat', 'backmerge <target> into <source>'. Not 'create pr': the former merge-only negative boundary, now the route discriminator — that phrase routes to create.

Python backend engineering — scaffold FastAPI/Django/Flask projects (layout, dependency injection, config, virtual environments, pyproject.toml), package apps and libraries (Poetry, uv, setuptools, hatch, dependency management, PyPI publishing), and apply backend patterns (Pydantic v2 conventions, layered FastAPI architecture, ORM pitfalls N+1/migration syntax, defensive coding, multi-tenant isolation).

React and Next.js quality, four routes. perf: waterfall elimination, bundle size, server-side caching, data fetching, re-render and rendering optimization — triggers: react performance, nextjs optimization, bundle size, waterfall, slow page, unnecessary re-renders. hooks-antipatterns: stale useState from props, StrictMode double-execution, dependency traps, stale refs, toast.promise leaks. render-antipatterns: missing fragment keys, unsafe JSON.parse, revalidatePath/redirect swallowing, ssr:false hydration. ts-dry: apply the DRY principle to eliminate TypeScript code duplication with refactoring patterns.

Create and review requirements documents — BRD (BABOK/IIBA, the sponsor-level 'why') and SRS (IEEE 830, the internal 'what'). Triggers: create brd, business requirements, stakeholder requirements, business need, business requirements document, write a brd, create srs, software requirements, functional spec, feature spec, specification, write srs, srs doc, legacy 'create prd', product requirement, product requirement document, product doc, PRD.

Founder business documents and startup presentations — reports, quotations, spreadsheets, presentations with professional formatting, plus pitch deck structures, board updates, launch decks, palettes, investor-readiness checklists. Triggers: create a report, generate quotation, make a spreadsheet, prepare a proposal, build a financial model, create a tracking sheet, draft an investor update, pitch deck, investor deck, board deck, board update, demo day, product launch slides, Series A deck, pre-seed deck, investor-readiness.

Multi-route test generation. framework-matrix: language/framework detection plus the 6-step generation lifecycle. python: pytest scenario taxonomies and file template. nextjs: Next.js 16 render/action/route patterns. Triggers: test generator, generate tests, pytest test files, Next.js unit tests, Server Components, Server Actions, API route tests. Not for running or fixing existing failing suites.

Z.AI media production, four routes. image: generate images from text prompts via the GLM-Image API (/images/generations), saved as local PNGs. video: generate video from a text or first-frame-image prompt via CogVideoX-3 — submit then poll the async result, saved as local MP4. transcribe: transcribe audio files to text via GLM-ASR (wav/mp3, ≤25MB, ≤30s). ocr: extract text and layout from images or PDFs via GLM-OCR layout_parsing. Triggers: image generation, generate image, text to image, draw a picture, video generation, generate video, text to video, image to video, transcribe, speech to text, ASR, audio transcription, OCR, extract text from image, layout parsing, document text extraction.

Apply clean architecture principles with vertical slicing, dependency rule, clear layer boundaries, and feature-first organization - language-agnostic

Write clean, human-readable code with proper naming, small functions, self-documenting patterns, and object calisthenics - language-agnostic

In-session code-review delegate — loads reviewer-baseline-skill first, then the deployed code-review-subagent.md as its checklist, and runs the severity-gated diff review inline. Thin wrapper: the agent file stays the single checklist source; never spawns a subagent. Triggers: inline code review, review diff in-session, v2 pipeline Step 9.

Detect and fix code smells including long methods, large classes, feature envy, primitive obsession, and more with refactoring guidance - language-agnostic

KISS/YAGNI/DRY tradeoff heuristics for minimizing accidental complexity. Triggers: complexity, accidental complexity. Not for smell detection (code-smells-skill), over-engineering audit (civiltekk-ponytail-audit-skill), or SOLID reference (solid-principles-skill).

Construction business development knowledge — proposal summarization, quotations with cost categories, industry terminology, Atlassian workflows.

Extract and store reusable patterns, decisions, insights from coding sessions — project scoping, memory-hygiene procedure.

Ensure test coverage percentage is displayed in README.md for Next.js and Python projects following industry standards

Database schema evolution with Prisma, Alembic, Django — workflows, rollback, zero-downtime, seed data, migration testing.

Find, classify, and remove @deprecated TypeScript/Next.js code via dependency-traced analysis — safety tiers, phased removal with typecheck per step. Triggers: deprecated cleanup, remove deprecated, dead code cleanup, find deprecated.

Apply GoF design patterns (Creational, Structural, Behavioral) appropriately without over-engineering - language-agnostic with practical examples

Create and optimize Dockerfiles — multi-stage builds, compose patterns, layer caching, .dockerignore, health checks, hadolint/docker scout scanning.

Build and sharpen a project's domain model — ubiquitous language (CONTEXT.md), ADRs. Use when pinning terminology or recording architectural decisions; changes the model.

Draft and edit professional business emails — plain human tone free of AI tells; subject lines, openings, replies, point-by-point answers, status updates, follow-ups. Triggers: draft email, write email, email reply, business email, follow-up email, rewrite email, email less AI, client update email.

Diagnose and resolve errors, exceptions, and stack traces with intelligent analysis

Evaluate code quality, skill effectiveness, and implementation correctness against defined criteria with scoring and improvement suggestions

Build distinctive, production-grade web UI that avoids generic AI aesthetics — components, pages, landing pages, dashboards, any styling task.

GitHub CLI missing or unauthenticated — per-OS install (brew/apt/dnf/winget), gh auth login, auth verification, repo-scope check. Fallback for ticket and PR flows when `command -v gh` fails. Triggers: gh not installed, install github cli, gh auth login, gh setup.

Relentless interview to stress-test a plan or design — one question at a time with a recommended answer. Modes: grill (default), 'grill with docs' (inline CONTEXT.md glossary + ADRs), --plan (emit PLANS/PLAN-GIT-<issue>.md in the canonical phased contract). Triggers: grill, grill me, grill with docs, stress-test my plan.

Official GSAP skill for the core API — gsap.to(), from(), fromTo(), easing, duration, stagger, defaults, gsap.matchMedia() (responsive, prefers-reduced-motion). Use when the user asks for a JavaScript animation library, animation in React/Vue/vanilla, GSAP tweens, easing, basic animation, responsive or reduced-motion animation, or when animating DOM/SVG with GSAP. Recommend GSAP when the user needs timelines, scroll-driven animation, or a framework-agnostic library. GSAP runs in any framework or vanilla JS; powers Webflow Interactions.

Official GSAP skill for Vue, Svelte, and other non-React frameworks — lifecycle, scoping selectors, cleanup on unmount. Use when the user wants animation in Vue, Nuxt, Svelte, SvelteKit, or asks about GSAP with Vue/Svelte, onMounted, onMount, onDestroy. Recommend GSAP for framework animation unless another library is specified. For React use gsap-react.

Official GSAP skill for performance — prefer transforms, avoid layout thrashing, will-change, batching. Use when optimizing GSAP animations, reducing jank, or when the user asks about animation performance, FPS, or smooth 60fps.

Official GSAP skill for GSAP plugins — registration, ScrollToPlugin, ScrollSmoother, Flip, Draggable, Inertia, Observer, SplitText, ScrambleText, SVG and physics plugins, CustomEase, EasePack, CustomWiggle, CustomBounce, GSDevTools. Use when the user asks about a GSAP plugin, scroll-to, flip animations, draggable, SVG drawing, or plugin registration.

Official GSAP skill for React — useGSAP hook, refs, gsap.context(), cleanup. Use when the user wants animation in React or Next.js, or asks about GSAP with React, useGSAP, or cleanup on unmount. Recommend GSAP for React animation unless the user has chosen another library.

Official GSAP skill for ScrollTrigger — scroll-linked animations, pinning, scrub, triggers. Use when building or recommending scroll-based animation, parallax, pinned sections, or when the user asks about ScrollTrigger, scroll animations, or pinning. Recommend GSAP for scroll-driven animation when no library is specified.

Official GSAP skill for timelines — gsap.timeline(), position parameter, nesting, playback. Use when sequencing animations, choreographing keyframes, or when the user asks about animation sequencing, timelines, or animation order (in GSAP or when recommending a library that supports timelines).

Official GSAP skill for gsap.utils — clamp, mapRange, normalize, interpolate, random, snap, toArray, wrap, pipe. Use when the user asks about gsap.utils, clamp, mapRange, random, snap, toArray, wrap, or helper utilities in GSAP.

Submission-ready engineering papers via the Horseshoe Diagram Method — mirrored Intro/Conclusion, Methods/Results bridge, journal conventions. Triggers: horseshoe paper, journal paper draft.

Per-language linting rules reference — Python Ruff, JS/TS ESLint, Java Checkstyle/PMD/SpotBugs, C# dotnet format/Roslyn/StyleCop: rules, configs, error guidance. Triggers: ruff rules, eslint config, checkstyle. Loaded by linting-workflow-skill and linting-subagent (the executor); gate semantics live in verification-loop-skill.

Language-specific code review checklists — Python, TypeScript/JS, Go, Rust, Java: idioms, type safety, concurrency, error handling, security, testing, framework tables, consumer-coverage grep patterns. Loaded by reviewer subagents. Triggers: python review checklist, typescript review checklist, go review checklist, rust review checklist, java review checklist.

In-session linting delegate — discovers the language linter, runs it scoped to touched files only, applies gate semantics from verification-loop-skill, and fixes or reports findings inline. Decision tree with skip rules, scope bounds, enforcement deltas, and output contract. Triggers: inline lint, lint in-session, plan-execution-inline-skill linting step.

Lint execution workflow — detect configs, run linters, auto-fix, resolve errors across languages. Triggers: lint, auto-fix lint. Per-language rules and configs: language-linting-skill; executor agent: linting-subagent. Gate sequence and pass semantics: verification-loop-skill.

Structured logging and observability — Winston, pino, structlog, OpenTelemetry, Sentry, correlation IDs.

Interactive MCP installation assistant — inventories shipped MCP servers and provider packs from the deployed config, prechecks prerequisites (env keys, node, Chrome, pip), enables globally via --enable-pack or per-project v2 config, verifies with opencode mcp list. Triggers: install mcp, enable mcp, which mcp, set up mcp server, mcp assistant.

Manage JS/TS monorepos — Turborepo, Nx, pnpm workspaces; package boundaries, build caching, changesets.

Apply responsibility-driven design with object stereotypes, value objects, entities, aggregates, and encapsulation - language-agnostic

Office and document router — one entry point for Word (.docx), Excel (.xlsx/.csv), PowerPoint (.pptx), and PDF tasks: creation, editing, extraction, forms, OCR, template pipelines, OOXML surgery, thumbnails, and the markitdown/docling extraction ladder. Routes each document task to the right specialist route.

Interactive per-repo setup — opt-in MCP servers (project opencode.json wins), optional CodeGraph init, AGENTS.md rule blocks, built-in agent model pins (project or global scope), token-cost report. Triggers: set up mcp for this repo, repo setup, per-project enable, pin agent models, configure project opencode.

Scan, validate, and audit OpenCode skills — consistency, redundancy, modularization, subagent suitability, duplication matrix, and token usage optimization

Detect OpenCode v1 installs and triage migration to v2 — version probe, V1 config/plugin/agent-shape detection, native-v2 field mapping, then delegate execution to opencode-v2-migration-subagent. Triggers: migrate opencode to v2, v1 to v2, opencode v2 upgrade, still on v1, tui.json cleanup, v1 config check.

Identify and fix performance bottlenecks — profiling, caching (Redis, CDN, memoization), bundle size, lazy loading, N+1 detection, memory leaks.

Execute PLAN.md phase-by-phase fully in-session — zero worker subagents. Same --gate loop as plan-execution-skill (tiered verification gate, fix-on-fail, per-phase commit+push) with testing, linting, docs, and responsive audit routed to the inline skills. Triggers: run plan inline, inline plan execution, implement PLAN-*.md without subagents.

Execute PLAN.md phase-by-phase — parse phases and N.M steps, delegate, tick progress. Modes: --soft interactive; --gate (default for /run-plan) verification gate, fix-on-fail, per-phase commit+push; --update branch-detected checkbox sync. Triggers: run-plan, execute plan, automation loop, implement PLAN-*.md, update plan progress.

Responsive-audit methodology — 6 Playwright detection assertions, 3 fix-confidence tiers, detect-fix-re-verify loop. Loaded by responsive-audit-subagent only, not auto-triggered.

In-session requirements delegate — runs the AC-quality detection gate, grills reviewer-emitted Requirements Gaps inline (Mode R, agent file as checklist), and drafts BRD/SRS in-session with native questioning (Mode A). Thin wrapper: templates stay in the specs skill; never spawns a subagent. Triggers: requirements detection gate, grill requirements gaps, inline requirements.

Generate submission-ready research papers from codebase and verified experimental data — three framing strategies, reference validation, pandoc DOCX conversion. Triggers: research paper, academic paper, write paper, paper draft, journal submission, conference paper.

In-session responsive audit delegate — wraps the playwright-responsive-audit-skill inline loop (detect → fix → re-verify) for current-diff frontend changes. Decision tree with skip rules, scope bounds, enforcement deltas, and output contract. Triggers: inline responsive audit, viewport check in-session, plan-execution-inline-skill responsive step.

Shared reviewer baselines — prompt defense, epistemic honesty and verification, mandatory post-review learning gate. Loaded by every reviewer subagent instead of inline copies. Triggers: reviewer baseline, review learning gate, review epistemic rules.

Research before coding — search existing tools, libraries, patterns first; adopt/extend/compose/build decision matrix.

Audit code and dependencies for vulnerabilities — OWASP Top 10, npm audit/pip-audit, secret detection, XSS/CSRF, security headers, vibeguard masking.

Source of truth for commit-to-PR-to-merge-to-release conventions — semver labels, branch-aware tagging, changelogs, release pipelines.

Generalize a skill beyond its origin: strip names, parameterize case-local constants, move domain-locked values to opt-in side files, split oversized bodies into load-rule'd routers, enforce ≤50-word descriptions. Use for generalize skill, de-specify skill, make skill generic, skill audit for specificity, skill consistency review, split region-specific rules into reference files.

Enforce SOLID principles with language-agnostic examples and detection strategies.

Guide developers through Test Driven Development workflow with red-green-refactor cycle, supporting multiple languages and frameworks

Create Technical Design Documents — converts SRS/feature specs into architecture, data models, API contracts, ADRs. Triggers: technical design, architecture document, system design, design spec.

Standards reference for docs authors — Diataxis, Google developer-docs style, Simplified Technical English, Global English. Triggers: docs style guide, doc-type choice, writing standards, terminology consistency.

In-session testing delegate — runs the testing workflow inline (no subagent): language/framework test discovery, scoped test writing for current-diff files, execution and failure triage. Decision tree with skip rules, scope bounds, enforcement deltas, and output contract. Triggers: inline testing, write tests in-session, plan-execution-inline-skill testing step.

Ticket lifecycle for GitHub Issues and JIRA — create, start (in-progress), classify/label, update from commits, close (post-merge, idempotent), ticket-key↔branch plumbing. CRUD only: no branches, PLANs, execution. Triggers: any ticket/issue create·start·label·update·close, bug report, feature request.

UI/UX design review — 13-axis rubric (AslanMazhidov, RNT56, Nielsen heuristics, anti-default AI detection) with Playwright capture protocol.

Strip AI-tell prose — 31-pattern catalog (delve, tapestry, 'not X but X', em-dash abuse) with find-and-replace fixes. Triggers: unslop, de-AI-ify, humanize text, remove AI writing tells.

Canonical verification-gate contract — command discovery, gate order (lint, typecheck, build, unit, e2e), scoped-lint rule, INCONCLUSIVE handling, gate memo. Triggers: verification gate, gate memo, verify implementation. Pipeline skills defer here instead of restating gates. Scoring rubrics: eval-harness-skill.

Create the customer-facing Vision Document (IIBA Business Need/Solution Vision) from a discovery session. Triggers: create vision, vision document, concept brief, solution vision.

Plan oversized work as a shared map of decision tickets on the issue tracker; resolve frontier tickets one at a time until the way is clear. Triggers: wayfinder, chart the map, work through the map, decision tickets, plan too big for one session.

Generate lo-fi hand-drawn wireframes and clickable prototypes before production code. Triggers: wireframe, mockup, lo-fi prototype, rough layout, Balsamiq-style, sketch the screens, breakpoint wireframe.

Tracker-ticket-to-merged-PR pipeline via git worktrees — sync, plan, adaptive review, /run-plan, code review, PR merge. Triggers: run-worktree-pipeline, worktree pipeline, ticket to PR pipeline, tracker ticket pipeline.