Claude Skills by ModernNomad-98

TEMPLATE ONLY — not a real skill and never invoked. Copy this directory to .claude/skills/<your-skill-name>/, rename it, set `name` to match the new directory, and rewrite every section against docs/skill-generation-standard.md. Write the real description to the Portability contract: front-load what the skill DOES in the first ~90 characters, keep it ONE strict-YAML-valid line (single-quote it like this one and double internal apostrophes, ''like this''), parsed value under 1024 characters.

Design AND read A/B and controlled experiments honestly — a falsifiable hypothesis, ONE primary metric plus guardrails, the randomization unit (sticky, non-contaminating), a sample-size/power calculation from a minimum detectable effect, and a pre-set duration that respects weekly cycles and forbids peeking-to-significance; then the READOUT — statistical vs practical significance with confidence intervals, the peeking/early-stopping trap, multiple-comparison and segment-fishing caveats, sampl...

Design the accessibility verification harness — WCAG-oriented coverage of keyboard operability (tab order, visible focus, no traps), accessible names/labels/roles, focus management in dialogs and route changes, color contrast, and screen-reader expectations (announcements, landmarks, live regions) — as BOTH automated tooling (axe-style scans in component and E2E layers, contrast checks in CI) AND a manual checklist for what automation cannot judge. Findings map to WCAG criteria with severity ...

Design the internal ops/support/superadmin CONSOLE for a multi-tenant SaaS — the surface operators act THROUGH: cross-tenant read/write with MANDATORY audit on every action, impersonation/support-mode with hard boundaries + consent + audit, least-privilege admin tiers (view-ops / write-ops / superadmin), break-glass elevation (time-boxed, approved, auto-expiring, logged), and the operator control-plane (health, manual failover/retry, data-repair). Produces the console architecture, the admin-...

Manage a CORPUS of Architecture Decision Records over time — the longitudinal discipline atop adr-writer (which authors ONE ADR): numbering and the ADR index/log, the status lifecycle (proposed → accepted → deprecated → superseded), bidirectional superseding links so a new decision points at the one it replaces and vice versa, detecting and resolving contradictions between ADRs, deciding when a change is a NEW ADR vs an amendment, and keeping the decision history navigable and trustworthy as ...

Write an Architecture Decision Record for a significant technical choice — context, the decision itself, alternatives genuinely considered, consequences (good and bad), operational impact, a rollback/reversal plan, and a review date. Use when a decision has just been made and needs durable recording, when asked to "write an ADR" or "document why we chose X", or when architecture-designer hands over an ADR draft for completion. Also use to record a rejected option or a superseded decision. Do ...

MANUAL-ONLY; never auto-invoke. Define the standing authorization matrix for AI agents — which actions an agent may take autonomously (read, plan, edit, test, commit, push feature branches, open PRs) and which ALWAYS require a named human decision (merging to protected branches, arming auto-merge — never agent-armed, deploys, releases, production data, secrets, history rewrites) — in a deny-by-default action × context matrix with approval scope/expiry semantics; the policy human-approval-boun...

Review whether a multi-agent system contains failure and drift (OWASP Agentic ASI08+ASI10 merged) — cascade half (ASI08): blast-radius isolation, bounded trust of upstream outputs, circuit breakers on inter-agent calls, checkpoints/rollback in multi-step pipelines, retry-storm and fan-out limits; rogue half (ASI10): drift detection against a behavioral baseline, agent inventory/lifecycle (no orphaned or shadow agents), and kill switches that sever authority (credentials revoked, not just proc...

MANUAL-ONLY; never auto-invoke. Recover from a broken working state — failed or interrupted runs, dirty or conflicted trees, partial or wrong commits, broken branches, blocked permissions — without losing work. Invoke explicitly when an agent session ended badly or git state looks wrong. Diagnoses read-only first, preserves everything (rescue branch or labeled stash, untracked files included) before changing anything, and never runs destructive cleanup (reset --hard, clean, force-push, branch...

MANUAL-ONLY; never auto-invoke. Defend an AI agent''s goal and plan integrity against goal hijack (OWASP Agentic ASI01) — pin the authorized goal at task start as a record only the authorizing principal can change, trace every step and sub-goal of a multi-step run back to the pinned goal, detect deviation (scope expansion, unrelated targets, substituted objectives) at plan checkpoints, and re-ground or halt on drift. Covers hijack via any untrusted channel: injected content, poisoned tool out...

Audit whether an AI-assisted change actually followed governance discipline — classification and scope lock, human approval at each crossed boundary, merge/deploy authority (WHO merged, who armed auto-merge), real validation evidence, security review where required, closeout completeness (incl. intentionally-not-done), and governed memory/doc updates. Per-control verdicts PASS/FAIL/UNVERIFIABLE cite primary evidence; closeout claims are cross-checked, never trusted; missing evidence is never ...

Design the governed operating environment an AI agent runs inside — the harness: ONE server-side mediation point every model/tool call crosses; identity from credentials (never from model-supplied payload), propagated; a deny-by-default pre-flight ladder (authenticate → authorize → entitlement → budget → input policy) BEFORE the model runs, each rung fail-closed; a CLOSED tool/provider registry (unknown capability fails, never improvised); instructions as server-side versioned artifacts no un...

Review agent identities and privilege paths (OWASP Agentic ASI03) — verify every agent has a distinct least-privilege identity (no shared god service account), credentials are task- and time-scoped, delegation chains (user → agent → sub-agent → tool) attenuate authority and never amplify it, confused-deputy paths are closed (agent privilege exploited by a lower-privilege requester), and every action attributes to both the human principal and the acting agent. Complements secrets-identity-hard...

MANUAL-ONLY; never auto-invoke. Audit and align agent instruction files across tools — CLAUDE.md (root, nested, local), AGENTS.md, Cursor rules, GitHub Copilot instructions, Windsurf and Cline rules, and similar — into one consistent source of truth. Invoke explicitly when instruction files conflict, duplicate each other, or have drifted, or after adopting a new AI tool. Produces an inventory, a conflict and duplication matrix, and a consolidation proposal with a rule-preservation diff; edits...

MANUAL-ONLY; never auto-invoke. Govern persistent agent memory as a curated, verified artifact — write rules (only confirmed durable facts with provenance and absolute dates; never secrets, credentials, tokens, or PII; never facts the repo itself records), trust rules (memory is a lead, not truth — reconcile remembered repo/PR/branch state against live git and gh output before acting on it), and hygiene (dedupe, correct stale entries, delete wrong ones, index consistency). Invoke explicitly t...

Run at the start of any repository or coding task, before reading or writing code. Verifies the working directory is the intended repo (git remote, landmark files), reads project instructions, status, and architecture docs, and separates verified facts from assumptions and missing information. Use when starting work in a repo, when told to cd into a path and build something there, or when resuming a session whose context may be stale. Halts and asks instead of building when the location or re...

Design or review least-privilege tool/function access for an LLM agent, containing excessive agency and tool misuse (OWASP LLM06 + ASI02) — build the per-tool permission matrix (side effects, blast radius, who it runs as), validate arguments against a schema before execution, run tools with the CALLING USER's authority (not a service account's), gate irreversible/high-impact actions behind human approval, map tool-chain abuse paths where one tool's output drives another's arguments, and class...

Design an agentic loop's shape and bounds — single-shot vs agentic as an EXPLICIT up-front decision (never default to a loop); clamped iteration ceilings; TYPED retryability (a policy rejection is TERMINAL, never retried; a transient failure is retried once on IDENTICAL input (reproducibility key) to classify flake vs deterministic); honest terminal states (an empty result is a legitimate stop, never forced into fabricated output); plan-act-observe-reflect with a defined stop. A loop needs an...

Produce the end-of-task closeout report — what changed, what was intentionally NOT done or was omitted (always a dedicated section, "None" written explicitly when empty), files touched, tests and validation actually run with real results, evidence, risks, skipped checks, and the recommended next action. Use when finishing a task, handing off work, opening or closing a PR, or when asked what was actually done. Scope reductions must be disclosed, never silent.

Design cost and consumption guardrails for an LLM feature covering both spend and unbounded-consumption abuse (OWASP LLM10 — denial-of-service and denial-of-wallet) — per-request token caps (input and output), per-user/tenant/plan budgets and cost-aware (token-based, not request-count) rate limits, model selection by task, concurrency and queue limits, loop/recursion bounds for agents, input-size limits, a fail-closed spend kill switch and degraded-mode fallback, and cost/usage telemetry with...

MANUAL-ONLY; never auto-invoke. Design and run the evaluation harness for an LLM feature — a versioned dataset (representative, adversarial/red-team, and regression cases), graders per dimension (task quality, schema adherence, safety/refusal, groundedness/hallucination, injection resistance, latency, cost), pass/fail thresholds, and a CI gate that blocks a prompt/model/retrieval/provider change on regression. Absorbs the AI security test harness: injection, jailbreak, data-exfiltration, and ...

Review the governance and risk posture of an AI feature — classify its risk tier by impact (who is affected, reversibility, autonomy, sensitive-data and rights exposure), assign accountable human ownership, require appropriate human oversight (advisory / human-in-the-loop / human-on-the-loop) matched to the tier, check user-facing AI disclosure and consent, data-use and retention posture, a model/feature card documenting intended use and known limits, and a map from applicable obligations (EU...

Operationalize the NIST AI RMF as the org''s AI risk program — the four Core functions (GOVERN cross-cutting, MAP context and risks, MEASURE analysis and tracking, MANAGE prioritization and response) applied across the AI lifecycle: design, development, deployment, operation, decommission. The AI RMF 1.0 is VOLUNTARY (released 2023-01-26, under revision per NIST; Generative AI Profile NIST-AI-600-1) — a risk method, NOT a certification target; the certifiable wrapper is its companion iso-4200...

Design controls against LLM misinformation and overreliance (OWASP LLM09) — require grounding (answer from retrieved/authoritative sources, not model memory, for factual claims), verifiable citations checked to actually support the claim, calibrated uncertainty and refusal-to-answer when evidence is thin, validation of consequential facts before they drive a decision or action, and UX that signals confidence and limits so users don't over-trust. Covers package/API hallucination (recommending ...

MANUAL-ONLY; never auto-invoke. Design the centralized model-routing layer all AI calls flow through — one internal interface in front of every provider/model so credentials live server-side only (never in the client bundle), routing picks the model by task/cost/availability, per-call telemetry is emitted, and budgets/rate limits from ai-cost-guardrail-designer are enforced at the choke point, with failure handled by retries/backoff, provider fallback, degraded responses, and a kill switch. C...

Define the end-to-end operating model for AI-assisted software work — the named stages (context, classify, plan, implement, validate, review, merge, close, learn), the authority holder at each (human, agent, or agent-with-approval), each stage's entry/exit gate, and its enforcing discipline skill — composing the Phase 1 pack plus agent-authorization-matrix, agent-memory-governance, and agent-governance-audit into one contract. Use when asked how humans and AI agents should plan, build, review...

Build the AI-specific threat model for an LLM feature, RAG pipeline, or agent BEFORE it ships — inventory AI assets (prompts, models, vector stores, tools, credentials) and trust boundaries where user input, retrieved documents, tool outputs, and model outputs are all untrusted by default; enumerate threats per boundary against the OWASP LLM Top 10 (injection, disclosure, poisoning, prompt leakage, excessive agency, unbounded consumption); write abuse cases from attacker behavior; rank risks ...

Design contract tests that VERIFY implementations against API, command, provider, webhook, and edge-function contracts WITHOUT UI tests — request/response schema validation (shape, types, required fields, error envelope), provider-side and consumer-side roles, versioning checks, and backward-compatibility gates (additive vs breaking diff rules) wired into CI. Contract = does the shape/version hold; not the business flow behind it. Produces the verification design and hands implementation off....

Design GENERATED API reference documentation — reference produced from the API's source of truth (OpenAPI/GraphQL schema, type annotations, docstrings) so it can't drift from the implementation, the split between what to auto-generate (the exhaustive reference) and what to hand-write (guides, examples, overviews), enriching the source of truth so descriptions/examples are single-sourced, the generation toolchain and where it slots into the docs pipeline, try-it/auth/error docs, and versioning...

Design external API and event contracts for a multi-tenant SaaS — tenant context derived from credentials (never client-supplied tenant ids on data paths), resource/route conventions, versioning and deprecation policy with sunset windows, idempotency for mutations, per-tenant and per-plan rate limits, and webhook/event-feed contracts (versioned envelope schemas, tenant-scoped subscriptions, at-least-once delivery with retries, signing and replay protection). Produces API contract conventions,...

MANUAL-ONLY; never auto-invoke. Implement a specific, already-decided application-security control in code — input validation at a boundary, output encoding, parameterized queries, authorization checks on a resource, secure session/cookie flags, safe file handling, SSRF/redirect allowlists, or a security fix for a known finding — test-first, with the control proven by a failing-then-passing negative test and no scope creep. Use when a threat model, security review, or the user has NAMED the c...

Advise on the architecture STYLE/paradigm for what someone is building — monolith, modular monolith, microservices, event-driven, serverless, service-oriented, or a hybrid — with honest, case-specific tradeoffs and a reasoned recommendation. Interview the need FIRST (domain, load, team size and operational maturity, constraints, scaling/change, consistency/latency) before advising; lay out only the genuinely relevant candidates for THIS situation (not a textbook dump); give case-specific pros...

Design or redesign system structure — components, dependencies, data ownership, integration points — grounded in an inspection of the CURRENT architecture first, never from an imagined one. Produces a component map, dependency map, coupling/cohesion risk list, data-ownership map, tradeoff analysis, an ADR draft for the key decision, and an incremental migration plan from current to target state. Use when asked how a system or feature should be structured, whether something should be a module ...

Design a durable, tenant-scoped audit log system — an audit event taxonomy (authentication events, access-control changes, data access and exports, admin/support actions, security events, billing changes), a record schema (actor, tenant, action, target, outcome, correlation id, timestamp), integrity guarantees (append-only, tamper-evidence, an explicit audit-write-failure policy), retention and redaction rules, tenant-scoped access to audit data, and negative tests proving audit writes cannot...

Diagnose and design the fix for the "change didn''t take effect" access-bug class — a removed user still sees the data, a revoked role still works, logout doesn''t end the session, a plan change still shows the old tier, a deleted item stays visible: inventory every place the old authority survives (server session records, JWT/token claims until expiry, client-side stores and data caches, server/CDN caches, live realtime subscriptions, database session context, search indexes, signed URLs), l...

Design the authorization model for a multi-tenant SaaS as an explicit roles × permissions × resources matrix — platform roles vs tenant roles, object-level authorization rules for every tenant-owned resource, an enforcement-point map (UI, API, service, background job, integration), deny-by-default posture, impersonation/support-access rules, and a negative-test plan proving forbidden actions actually fail. Includes an additive migration and rollback path for role changes. Use when designing o...

Map a DECIDED logical architecture onto provider-idiomatic AWS for a multi-tenant SaaS — account topology (Organizations, SCPs), identity (IAM roles, Cognito, OIDC federation), network (VPCs, PrivateLink, ALB/CloudFront with WAF), Secrets Manager/KMS, data with an explicit tenant-isolation strategy per store (Aurora/RDS, DynamoDB, S3), compute chosen by team maturity (ECS/Fargate, EKS, Lambda), messaging (SQS, SNS, EventBridge), CloudWatch/CloudTrail observability, Security Hub/GuardDuty/Insp...

Map a DECIDED logical architecture onto provider-idiomatic Azure for a multi-tenant SaaS — subscription layout, identity (Entra ID, managed identities, OIDC federation), network (VNets, Private Link, Front Door), Key Vault, data with an explicit tenant-isolation strategy per store (Azure SQL, Cosmos DB, Blob), compute chosen by team maturity (App Service, Container Apps, AKS, Functions), messaging (Service Bus, Event Grid), Azure Monitor observability, Azure Policy + Defender for Cloud/Sentin...

Design the async job/worker EXECUTION model for a multi-tenant SaaS — offload slow work off the request path: worker pools, scheduled/cron jobs, job idempotency + resumability/checkpointing, retry with backoff, a job dead-letter queue, visibility timeouts, and per-tenant fairness so one tenant''s flood cannot starve others. Produces the job catalog, the execution/retry/DLQ contract, and a fairness + scaling plan. Use when moving slow work out of the request, adding scheduled jobs, or when job...

Decide what gets cached, where, and how it stays correct — candidate analysis first (read/write ratio, miss cost, staleness tolerance stated per item), layer choice (HTTP/CDN edge, in-process, distributed cache, DB materialization), invalidation designed BEFORE the cache ships (TTL, event-driven purge with backstop, write-through/behind — with the consistency envelope each buys), key design where tenant scoping is a correctness boundary (tenant-qualified keys; a NEW cache store's isolation de...

Design cell-based (blast-radius) partitioning for a multi-tenant SaaS — a cell is a self-contained stack subset (compute + data + cache + queue) that serves a subset of tenants end to end, so a failure, bad deploy, or noisy tenant is contained to ONE cell instead of the whole fleet: cell definition, tenant→cell mapping and placement policy, a thin cell-router, cell-by-cell deployment/canary, cross-cell/global concerns, and cell migration/rebalancing. SCALE-STAGE — most SaaS never needs it; re...

Classify a requested change before starting work — docs-only, UI/style, frontend logic, backend/API, schema/migration, RLS/security, cloud/IaC, AI/agentic behavior, QA/test-only, refactor, bug fix, or release — and map the class to its required validation level and approval path. Locks scope to the approved class and file set; mid-task scope growth forces reclassification. Use when starting any non-trivial change, when a task mixes change types, or when work is drifting beyond what was asked.

On a cadence, extract decisions, bugs, and backlog items that exist only in ephemeral AI-chat history into dated repo docs, then AUDIT every item against PR/source evidence — classified completed / partial / active / not-active / unknown with citations; a chat "done" caps at unknown until repo evidence upgrades it. Enforces the standing rule: do not rely on stale chat history — tracked repo docs are the working record. Produces the dated extraction doc, the per-item audit table, and survivors...

MANUAL-ONLY; never auto-invoke. Design the delivery pipeline and, on request, edit the pipeline definitions — the stage graph (lint, typecheck, build-once, unit, integration, security, E2E, artifact, deploy gates) with explicit merge-blocking semantics and a latency budget, CI secret governance (OIDC over stored credentials, job-scoped secrets, none to fork PRs), artifact provenance and cache trust, environment promotion with named-human gates, deployment strategies with rollback primitives, ...

MANUAL-ONLY; never auto-invoke. Execute a systematic interactive clickthrough of a RUNNING app — a route-by-route walkthrough plan covering navigation, forms (valid + invalid input), dialogs, permission-gated controls, empty/loading/error states, and destructive-action confirmations, then drive the UI through it, recording per-step observations with severity-rated defects and screenshot evidence at named checkpoints. One-session verification pass, not a permanent scripted suite. Use when aske...

Decide cloud platform, abstraction rung, and deployment posture cloud-neutrally — gather requirements FIRST (compliance/residency, latency/regions, availability, cost, team maturity, existing estate), shape the provider-neutral logical architecture, THEN pick the highest abstraction rung the workload tolerates (IaaS/VPS → container-PaaS → managed-Jamstack/SSR → Postgres-BaaS → edge/serverless → hyperscaler), a provider within it (modern managed platforms e.g. Vercel/Render/Supabase/Cloudflare...

Review an ACTUAL diff — a PR, branch delta, or staged/working changes obtained from git — and report findings by severity (blocker/major/minor/nit), each with file:line evidence and a concrete remediation. Covers correctness, security, performance, reliability, maintainability, test adequacy, and migration safety, and reads enough surrounding unchanged code to judge the change in context. Use when asked to review a diff, PR, branch, or commit. Do NOT use for behavior-preserving cleanup applic...

MANUAL-ONLY; never auto-invoke. Apply behavior-preserving simplifications to explicitly named code — remove dead code and needless indirection, flatten nesting, deduplicate, replace hand-rolled logic with idiomatic equivalents — with tests proving behavior unchanged (green on the same suite before AND after every move). Manual-only by design; it edits working code, so it must never auto-trigger mid-task. Use when the user explicitly asks to simplify, clean up, reduce complexity of, or make mo...

Design a single server-side-mediated write path — a command bus — for protected mutations in a multi-tenant SaaS: a command registry plus a per-command pipeline (validate payload → authenticate the actor from the token, never from client-supplied identity → authorize against policy → server-derive tenant/resource scope from trusted rows → idempotency → execute → emit audit + domain events → safe error envelope), plus the no-direct-client-writes invariant for protected actions. Produces the co...

Author the framework-agnostic common control set that ISO 27001, ISO 42001, and SOC 2 work all consumes — one catalog across access control, cryptography, change management, logging/monitoring, incident response, vendor management, and risk assessment, each control written ONCE with objective, owner, implementing mechanism, and evidence hook, then projected per framework instead of rebuilt per framework. Controls MAP to shipped implementations by name (authorization-matrix-designer, rls-polic...