AI Agent Workflow Collection

Design or audit a product accessibility conformance program spanning target standards, representative workflows, manual and assistive-technology testing, issue governance, release evidence, VPAT/ACR inputs, remediation, and exceptions. Use when the requested artifact is a durable conformance program, claim-evidence review, or release/procurement decision. Do not use for a single UI critique, a code-level accessibility fix, generic QA planning, or legal advice.

Design or audit a secure, humane account-access recovery protocol for lost credentials, passkeys, MFA devices, recovery codes, email or phone access, social login, enterprise SSO, compromised accounts, locked users, ownership disputes, and support-assisted recovery. Use when the primary artifact is recovery eligibility, proof, risk, channel, state, authority, containment, communication, appeal, and post-recovery protection. Use enterprise-access-governance-review for tenant/admin role lifecycle and product-abuse-risk-review for cross-product abuse controls.

Design or audit one complete in-product advertising system for apps, games, web, and content products: rewarded ads, interstitials, banners/native placements, offerwalls, mediation, payer suppression, reward authority, consent, age/territory modes, frequency and fatigue budgets, IAP/IAA cannibalization, fraud, experiments, and shutdown. Use when deciding ad formats, placements, rewards, provider portfolio, or ad economics; combine with an App or Game Blueprint only when whole-product coherence is also unresolved.

Design or audit how an agent system acquires, selects, compiles, remembers, invalidates, and hands off information, producing a Context and Memory Contract. Use for working-context assembly, episodic or semantic memory, procedural instructions, retrieval and ranking, long-session compaction, provenance and freshness, cross-agent handoff, memory privacy, or stale-context failures. Do not use for ordinary document retrieval, prompt wording alone, planning algorithms, or live work-ledger coordination.

No-human parallel development packets, change attribution, collision scopes, and machine-verifiable acceptance. Use when defining or running multi-agent software delivery without a human review dependency. Do not use for a small local edit or live Work Graph operations.

Apply agent-native product constraints for queryable typed state, schema-derived tools, recoverable workflows, delegation, durable context, and model/tool/agent-run traces with protected operator diagnostics. Use when designing a product, API, or workflow that agents must operate directly and recoverably. Do not use for ordinary automation, generic service observability, or product analytics without an agent-facing contract.

Design or audit the planning and control mechanism of an autonomous agent or multi-step agent workflow, producing an Agent Planning and Control Contract. Use when choosing or evaluating BDI, HTN, partial-order or causal-link planning, Plan-and-Execute, ReAct, receding-horizon control, bounded search, replanning, execution monitoring, or stop conditions. Do not use merely to plan one ordinary task, schedule a current work queue, design agent memory, or review a finished code candidate.

Improve a recurring agent-system failure through a causal Learning Change Record, controlled intervention, replay or holdout evidence, promotion criteria, and recurrence monitoring. Use when prompts, Skills, context assembly, tools, evals, coordination, or agent workflows repeatedly produce the same error, cost, delay, or weak outcome. Do not use for a one-off product-code bug, live incident response, ordinary task planning, or evaluation of one already-defined Skill candidate alone.

Design or audit one AI product behavior and launch contract covering the user job, AI necessity, model role, autonomy and action rights, data use, affected parties, failure and abuse, UX disclosure, fallback, cost and latency, support, evidence dependencies, and release boundaries. Use when an assistant, copilot, generator, classifier, recommender, or agent needs a product-risk dossier. Do not use for eval implementation, adversarial red teaming, security threat modeling, provider selection, a whole-product blueprint, or incident response.

Design or critically review a complete app as a coherent, agent-first App Design Blueprint: user job and promise, workflows and information architecture, first-use activation, capability portfolio, retention and durable ownership, commerce and social integration, feedback and review loops, responsive HTML5/mobile/desktop experience, desktop OS integration semantics, offline and sync, accessibility, localization, low-end performance, privacy, safety, autonomous construction, and validation. Use for a new consumer or productivity app, whole-app redesign, cross-platform app system, a bounded macOS/Windows integration design that must remain consistent with app state, or an audit of whether product capabilities form one scalable experience. Use specialist procedures for one payment, refund, notification, review, campaign, store submission, exact release artifact, or file-level interface artifact unless whole-app coherence is also being decided.

Deliver a substantial code-architecture refactor or migration to one declared target across one or many repositories. Use for whole-repository Capability-first DDD adoption, language/runtime rewrites, monolith restructuring, drop-in replacements, or architecture cutovers that require real code movement, semantic proof, authority change, and predecessor retirement. Do not use for a small local refactor, a documentation-only architecture review, or customer-data migration.

Autonomous scheduling: critical-path focus, work-conserving progress, complexity-gated resource-safe delegation, replanning, and honest evidence layers. Use when a non-trivial task should continue through multiple independent work lanes without repeated user direction. Do not use for a short direct answer, a few file reads, one command, or a tightly coupled local edit.

Prepare or audit the operating-metrics evidence packet behind a board or executive review: metric definitions, source reconciliation, period snapshots, cohort and segment cuts, variance bridges, forecast confidence, owner signoff, decisions, and action closure. Use when leadership needs board-ready operating truth. Do not use to invent company numbers, produce audited financial statements, design product analytics instrumentation, analyze one retention chart, or write investor persuasion detached from evidence.

Define and evaluate the causal effect of an intervention, exposure, policy, or treatment on an outcome using explicit estimands, causal assumptions, study design, identification, diagnostics, and sensitivity analysis. Use for questions such as whether X causes Y, what would happen if X changed, observational or quasi-experimental impact analysis, confounding, mediation, selection, or causal attribution. Produce a Causal Inference Record. Do not use for ordinary correlation reporting, isolated software root-cause debugging, generic system maps, or merely configuring an A/B test.

CI admission tiers, gate-portfolio discipline, affected-test selection, required checks, flake control, and preview policy. Use when designing, adding, removing, consolidating, or auditing checks that decide whether a candidate may merge, including custom lints and architecture tests. Do not use for debugging one unrelated test failure.

Design or diagnose CI runner capacity, queue/pickup latency, execution profiles, merge-queue backpressure, inventory uncertainty, and critical-path isolation. Use when CI jobs wait, runners disappear, queues saturate, or capacity policy must change. Do not use for a candidate failure caused by product code.

Pricing, packaging, entitlements, commercial ADRs, and market-positioning evidence. Use when making a durable revenue, packaging, licensing, or route-to-market decision. Do not use for implementing an already-decided checkout or billing contract.

Run a finite, high-coverage formal review of an exact candidate against a frozen acceptance contract, using independent perspectives and executable oracles, batching repairs, and closing without moving the goalposts. Use when review convergence itself is the job. Do not use for open-ended exploration, root-cause hypothesis generation, option selection, or one quick opinion.

Design or audit one customer-facing data import, onboarding migration, tenant move, system replacement, or bulk cutover from source discovery through mapping, transformation, validation, rehearsal, execution, reconciliation, rollback/forward-fix, customer communication, and acceptance. Use when the primary artifact is a repeatable migration contract for customer data and service continuity. Use architecture-convergence for a substantial one-or-many-repository code-architecture rewrite, product-recovery-contract-review for same-product backup and restore, and customer-success-operations-review for the ongoing post-migration value program.

Design or audit an outcome-led customer success operating system across segmentation, onboarding after technical migration, activation, adoption, health, risk intervention, education, business reviews, renewal and expansion signals, champion or stakeholder change, playbooks, automation, and product-learning handoffs. Use when the primary artifact is the recurring post-sale customer-value system for B2B SaaS, developer tools, platforms, or enterprise products. Use customer-data-migration-review for import/cutover execution, customer-support-operations for reactive support queues, and enterprise-contract-operations-review for contractual obligations and renewal mechanics.

Resolve or audit one customer-support case from verified facts through a humane response, remedy or compensation, protected-action handoff, legal or abuse escalation, follow-up, and closure evidence. Use when the independent artifact is one Customer Case Resolution Record containing a customer-ready reply when needed, including for a complaint, billing or refund request, trial request, goodwill or gift request, abusive contact, or legal threat. Use Customer Support Operations for the whole support system and Refund And Support Flow for the governing refund consequence state machine.

Design or audit one end-to-end customer-support operating system across self-service, intake, identity-safe diagnostics, taxonomy, queues, severity, routing, SLAs, grounded responses, autonomous resolution, specialist and incident handoffs, quality calibration, appeals, trust recovery, knowledge freshness, product feedback, metrics, and continuous improvement. Use when the whole support model is the independent artifact; use Refund And Support for one refund consequence state machine and do not invoke for one reply or a fixed QA rubric.

Design or audit one daily reward, check-in, streak, calendar, mission, repair, comeback, or return-ritual system for apps and games, including time semantics, progression gates, reward economy, notifications, paid repair/VIP/gacha interaction, abuse, fairness, recovery, experiments, and support. Use when the recurring artifact is a return-loop state machine; use Game/App Design for whole-product retention and Notification Strategy for channel delivery.

Design or audit a recurring Data Reliability Contract for data products and pipelines—datasets, streams, imports, transformations, warehouses, and search/operational projections—across freshness, completeness, semantic drift, reconciliation, trust state, repair, and consumer impact. Use when ongoing data fitness is the primary artifact. Do not use for one bounded customer/tenant migration, search relevance/ranking quality, a general provenance/custody graph, product analytics metrics, ledger truth, privacy, or active incident response alone.