AI Agent Workflow Collection

Gate deploy on a security audit and invariant tests across the top vuln classes — deploy is irreversible. Reach for this before any mainnet go-live.

Audit segregation of duties and chart-of-accounts hygiene before trusting the books. Reach for this on a controls or data-quality question.

Read note timeliness and medical-necessity completeness as one revenue-and-compliance control — operationally, never as a clinical judgment. Reach for this on a denial or audit-readiness question.

Make the backend survive its dependencies: timeout every outbound call, retry idempotent-only with exponential backoff + jitter, add circuit breakers and bulkheads, define a graceful-degradation mode, and design idempotent background workers with DLQs and backpressure.

Track flash and RAM (image, static, worst-case stack/heap) against the part's limits. Reach for this on any memory or footprint question.

Build the EGI-to-NOI bridge from gross potential rent and translate to value at a cap rate. Reach for this on an NOI or valuation question.

Build the duty-cycled current profile and compute average current and battery life. Reach for this on any battery-life or energy question.

Build Bulk API 2.0 ingest and query jobs for high-volume data movement. Use when loading or extracting large datasets over the REST Bulk API rather than row-by-row.

Hold quality in control and stop defects recurring: stand up SPC reading special vs common cause correctly, structure a nonconformance through containment into a real CAPA, build inspection/control plans and FMEA, and set the supplier-quality bar — prevention over detection over scrap.

Onboards a non-Claude-Code agent (GitHub Copilot CLI / Cursor / Aider / Codex / Devin) to this repo on session start. Use when this repo will be operated on by a coding agent that does NOT natively read CLAUDE.md. Mirrors what CLAUDE.md provides to Claude Code, but routes through AGENTS.md (which all major 2026 agents read) and surfaces tool-version floors + 2026 H1 failure-mode mitigations.

Compute the service absorption rate against total fixed overhead — the survival metric. Reach for this on a fixed-ops question.

Playbook for allocating a Claude context window across system prompt, retrieved documents, conversation history, and tool results — with token-budget formulas, the retrieve-vs-hold decision, and the compaction triggers that prevent context overflow. Owned by prompt-and-context-engineer.

Design and audit conversion-focused screens — funnel definition, friction-vs-trust balance, form field selection (every field is a cost), one CTA per screen discipline, social-proof placement, microcopy patterns, and the conversion-rate measurement plan. Reach for this skill when designing or reviewing a landing page, sign-up flow, checkout, or any "convert visitor → action" screen. Used by ux-designer (primary) + content-strategist.

Diagnose and improve Core Web Vitals (LCP / CLS / INP). Includes a Mermaid decision tree for root-cause triage across all three vitals + canonical fix-by-symptom map. Field data (CrUX / RUM) at p75 is the measurement target. Used by performance-engineer (primary).

For any script that produces a file the repo COMMITS and CI DIFFS — generated HTML, lockfiles, JSON snapshots, fixture files, documentation rendered from source. The output must be byte-identical regardless of who regenerated it or on what OS. This skill catches the specific failure modes that turn a freshness gate into a paper tiger that fails forever — OS-dependent path separators, locale-dependent encodings, nondeterministic ordering, drifting timestamps. Triggers on phrases like "freshness check failing", "regenerated locally but diff doesn't match", "works in CI not on Windows", "generator produces different output", and on any review of a script whose output is committed.

Design defensible comp bands tied to leveling and dated market data — set midpoints/spreads, compute compa-ratio and range penetration, surface over/under-band outliers. Reach for this on a banding or offer question.

Build retention into visit burden, schedule, and engagement to lower the ~30% dropout, instead of re-recruiting. Reach for this when dropout threatens the timeline.

Design a paved road that is the lowest-friction compliant option, with self-service actions and guardrails. Reach for this on a paved-road question.

Diagnose application-to-funded fall-out stage-by-stage and name the worst fallout stage — fix it before buying more apps. Reach for this on a pull-through question.

Locate a conversion problem by funnel stage — traffic, product page, cart, checkout — instead of reading the headline rate. Reach for this when conversion is low.

Draft the sustainability disclosure with its evidence trail, build the data controls, assess readiness for the target assurance level (limited vs reasonable), run a gap assessment against what an assurer will test, strip greenwashing, and prepare the auditor-liaison pack.

Evaluate channels on contribution and marginal ROI under a stated attribution model — not average ROI or lead count. Reach for this on a channel or budget question.

Tie fallout assumptions to locked-pipeline exposure and frame the lock/pipeline risk — route the hedge to the risk authority. Reach for this on a lock or rate-risk question.

Make a frontend fast to a budget: analyze and code-split the bundle by route, lazy-load heavy/below-the-fold, optimize images/fonts, minimize hydration (RSC/islands), kill render-blocking and request waterfalls, and tune the Core Web Vitals (LCP/INP/CLS) against field data.