AI Agent Workflow Collection

Web accessibility (a11y) discipline. Semantic HTML over div soup, ARIA only when semantic markup is insufficient, color contrast ≥4.5:1, focus order matches reading order, keyboard navigation, screen-reader-friendly forms, alt text discipline. Contains paste-ready component patterns, contrast tokens, and a focus-ring recipe. Use whenever building UI that users will navigate with a keyboard or screen reader.

Writing self-contained subagent briefs. Four-part structure (goal, context, constraints, return shape), token budgets, autonomy bounds, hand-the-question vs hand-the-command, anti-motivational filler. Contains ready-to-paste templates for investigation, implementation, and verification subagents. Use when an orchestrator agent needs to spawn a subagent and write its prompt.

REST and RPC discipline for backend APIs. Resource naming, canonical error shape, cursor pagination implementation, idempotency keys, status code decision logic, versioning. Contains worked TypeScript implementations of the patterns. Use when designing new endpoints or auditing existing ones.

Authentication and session discipline. argon2id password hashing, server-side sessions vs JWT, OAuth 2.1 + PKCE, MFA via TOTP / WebAuthn, password reset hygiene, anti-enumeration timing, rate-limited login. Contains ready-to-paste session middleware, password hash helpers, MFA enrollment, reset flow. Use when designing or auditing auth - the highest-leverage place for security bugs to ship.

Brand kit image generation - logo systems, color/type/grid boards, photography direction, applications. Six to twelve board set with a single visual register held throughout. Eight canonical registers (minimalist tech / editorial / dark-tech / luxury / cultural / gaming / developer-tool / consumer-app). Contains prompt template + 9 worked board prompts. Use when producing brand guideline boards, identity decks, logo concept sheets.

Hard mechanical UI register. Swiss typographic print + military terminal aesthetics. Rigid grids, extreme type contrast, sans+mono pair, utilitarian color, numbered sections, hairlines instead of shadows. Contains palettes, type stack, numbered-heading pattern, terminal-styled blocks. Use for data-heavy dashboards, technical portfolios, editorial-tech sites. Mutex with forge-minimalist and forge-soft.

Caddyfile discipline for production traffic. Global options, automatic HTTPS done right, security header pack, reverseproxy with health checks, structured logs, multi-tenant on-demand TLS with validation, rate limiting, fileserver hardening. Contains ready-to-paste Caddyfile + snippets. Use when writing or auditing a Caddyfile.

Citation discipline for AI-generated outputs. When to cite, what counts as a verifiable citation, URL + quote + file-path verification, marking inference vs fact, what NOT to cite. Contains paste-ready citation formats for markdown, RAG-with-chunks, and code references. Use whenever an LLM produces a fact-claim that someone might act on - RAG, research, code-with-references, legal/medical/financial outputs.

CLI design discipline. Unix-philosophy small composable commands, exit codes that mean things, subcommands with noun verb shape, --help that fits one screen, stdout for data + stderr for chatter, JSON output mode for piping, dry-run for destructive operations, config from flag > env > file. Contains paste-ready argument-parser patterns and a complete worked CLI. Use when designing a new CLI tool or auditing one.

Reviewing pull requests well. What to look at first, what to skip, severity prefixes, when to block vs approve, frame feedback as questions, never rewrite-as-review, reviewing AI-generated code specifically. Contains the severity-prefix vocabulary + a reviewer's mental checklist. Use whenever you are reviewing someone else's code.

Commit messages that survive git blame in five years. Subject under 72 chars imperative mood, body for the why not the what, no AI attribution, atomic commits, revert + migration + refactor patterns. Contains paste-ready templates. Use when authoring commits.

Production-grade Dockerfiles. Multi-stage builds, non-root execution, deterministic dependencies, sub-200MB final image, signal-handling exec-form CMD, build-time secret mounts, .dockerignore hygiene, language-specific defaults. Contains worked Dockerfiles for Node, Go, Rust, Python. Use whenever writing or auditing a Dockerfile.

Where to catch, where to throw, what to log, what to surface. Result types vs exceptions, sanitized client responses with requestid correlation, retries with backoff, idempotent recovery, panic on invariant violation. Contains ready-to-paste implementations. Use when writing or reviewing code paths that can fail.

Designing evals for LLM features. Golden datasets, rubric scoring, LLM-as-judge calibration, regression detection in CI, online A/B tests, cost+latency budgets, adversarial cases. Contains paste-ready eval harness with promptfoo / vitest, judge prompt template. Use when shipping an LLM feature to production, not for one-shot experimentation.

Feature-flag discipline. Flags as a deployment-decoupling tool, not a permanent toggle. Naming and ownership, default-off + kill-switch, targeting rules (% rollout / tenant cohort / user attribute), removal schedule baked in, dark-launches and gradual rollout. Contains paste-ready flag wrapper + a flag-cleanup audit. Use when introducing feature flags so they do not turn into permanent technical debt.

Premium frontend output without locking one visual style. Sharper layout, typography, motion, and spacing than the default agent baseline. Use for greenfield UI, landing pages, dashboards, marketing sites. Contains concrete failure-mode patterns to avoid, named palettes, named type stacks, and worked code examples.

Next.js 16 with React Server Components and Tailwind 4. Server-first by default, streaming with Suspense, server actions over API routes, app router conventions, async page params, prompt-cached metadata. Contains worked RSC page + server action + loading.tsx + error.tsx pattern. Use whenever writing Next.js 13+ (app router) code; pair with forge-frontend for visual taste.

Anti-truncation discipline. Forces complete code generation, bans // ... rest of the file placeholders and "omitted for brevity" prose. Specifies how to split cleanly when a file is genuinely too long for one response. Contains the banned-phrases list, the explicit-split format, and a worked example. Use when an agent keeps producing partial files the user has to ask to "continue.

GitHub Actions workflow discipline. SHA-pinned third-party actions, scoped permissions per job, OIDC federation over long-lived cloud credentials, dependency caching, concurrency cancellation on PR, no secrets in run: lines, no pullrequesttarget running fork code. Contains ready-to-paste lint+test+build+deploy workflow. Use when writing or auditing .github/workflows/.yml.

GraphQL schema design. Nullable defaults, connections (cursor pagination via Relay spec), errors as data not throws, DataLoader for N+1, persisted queries for caching, schema versioning via deprecation, no overfetch via fragments. Contains paste-ready schema patterns and a DataLoader recipe. Use when designing a GraphQL API.

Art direction for mobile app reference screens. One screen per image, portrait device aspect, subtle straight-on phone frame, status bar with realistic content, real-looking content not Lorem ipsum, palette locked across all screens. Contains the prompt template + 6 worked prompts for a full app flow. Use with image generation tools for iOS, Android, cross-platform mobile design.

Art direction for website reference comps generated by AI image tools. One horizontal image per section, no compressed mood boards, varied composition, palette locked across all images, second-read moment per page. Contains the prompt template + 8 worked prompts for a full landing-page set. Use with Midjourney, DALL-E, ChatGPT Images when producing references for landing pages and marketing sites.

Production-grade Kubernetes manifests. Pinned images, resource requests + memory limits (no CPU limits), securityContext (non-root, read-only root, drop all caps, RuntimeDefault seccomp), readiness probes, PodDisruptionBudgets, NetworkPolicy, external secrets, graceful shutdown. Contains ready-to-paste Deployment + Service + PDB + NetworkPolicy. Use when writing or auditing manifests for production traffic.

Streaming LLM output correctly. SSE setup, partial-token UTF-8 decoding, AbortController cancellation end-to-end (browser + server + upstream), mid-stream errors as events not HTTP status, append-only UI rendering, backpressure, time-to-first-token. Contains ready-to-paste SSE handler + browser fetch reader + cancellation chain. Use when building any LLM-facing UI that should feel responsive.