Project
HeartWire Agent Suite
18 domain-specialized Claude agents for engineering, finance, trading, and content ops
Summary
A personal bench of 18 specialist AI agents — each a Claude Code skill with an explicit role, activation trigger, persona, and capability list — spanning software engineering, hardware, finance, trading, legal, tutoring, and content.
Use case
Used daily across HeartWire engineering, coursework, trading research, and personal finance instead of one general-purpose assistant.
Who it's for
- personal
- engineers
Stack
- Claude Agent SDK
- Claude Code Skills
- MCP
- Markdown-defined personas
Case study
What it is
A personal suite of 18 domain-specialized AI agents, each built as a Claude Code skill — a bench of specialists to call into any project instead of one general-purpose assistant.
Used daily across HeartWire engineering, OmniFlex work, coursework, trading research, and personal finance, so the roster spans software, hardware, finance, legal, tutoring, and content.
How each agent works
Every agent is a Markdown file with a YAML frontmatter description Claude Code matches against the task at hand, then a body that defines five things: Role (what it is), Trigger (when it activates), Persona (how it communicates), Capabilities (what it can do), and Constraints (what it refuses to do).
The Senior SWE Agent, for example, is scoped to never invent APIs or config flags without flagging uncertainty, never commit or push unless explicitly asked, and never expand scope into product strategy — it hands that off to the Business Consultant Agent instead.
Several agents also carry a fixed output format so results are consistent run to run. The Senior SWE Agent always responds in the same structure: Intent, Approach (tradeoffs), Changes, Code, Tests, Risks/Debt, and How to verify — so a PR review reads the same way whether it's touching Supabase RLS or a React component.
Engineering & code
Senior SWE Agent — production code, PR review, and architecture across Python, TypeScript, React, Supabase, Vercel, and GCP/AWS. Activates for implementation, refactoring, debugging, or technical tradeoff decisions.
Backend Dev Agent — API design, database schema, auth flows, and serverless work on Supabase, Firebase, FastAPI, and Node. Activates for endpoints, RLS policies, webhooks, or auth/session work.
Architect Agent — system diagrams, ADRs, service boundaries, and React/React Native design guidance for cloud-native and agentic apps.
Code Auditor Agent — security and quality audits with severity-rated findings; scans for injection, XSS, IDOR, and SSRF risk across full-stack and agentic codebases.
Mechanical + CS Engineer — mechanical design, CAD/FEA, controls, mechatronics, and simulation, kept honest by separating the physical system from the model, the simulation, and the code that implements it.
Electrical + Hardware Engineer — circuit analysis, PCB design, embedded hardware, and signal integrity, worked requirements → first principles → implementation → verification.
Finance & trading
CPA/CFO Agent — bookkeeping hygiene, cash runway, budgets, and financial modeling. Runs a fixed cadence: weekly cash snapshot, monthly P&L narrative, runway/burn analysis, tax-deadline reminders, and a vendor/subscription audit — always flagging what needs an actual CPA's sign-off.
Investment Portfolio Agent and Portfolio Manager — crypto and equities research across five modes: holdings snapshot with allocation drift, a rebalance checklist against target policy, risk notes on concentration/correlation/liquidity, a watchlist research digest, and a logged decision journal entry.
Trading Agent — entry/exit thesis, risk/reward framing, and trade journals for crypto and equities; requires an explicit invalidation level before any setup counts as a thesis.
Tax Auditor Agent — expense categorization, deduction flags, 1099/crypto gain-loss organization, and quarterly-estimate prep, always separated from anything that requires a licensed preparer.
Legal, strategy & tutoring
Legal Agent — contract, NDA, and ToS red-flag review, IP ownership questions, and LLC structuring questions for a software/SaaS context; flags what needs actual counsel rather than bluffing case law.
Business Consultant Agent — PMF diagnosis, revenue modeling, SWOT, and GTM plans, biased toward concrete 7/30/90-day actions over slide decks.
Research Analyst — a seven-step pipeline: scope the question, build a query plan, intake sources, validate evidence, extract data into tables, synthesize findings, then propose next actions — built to keep every claim traceable to a source.
BME Tutor Agent and Math Tutor Agent — FGCU coursework support across biomechanics, signals, circuits, Calc III/DiffEq, and control systems, with every quantitative result reconstructable by hand on a blank exam page.
YouTube Agent — scripts, SEO titles, thumbnail briefs, and content calendars for long- and short-form video.
Example prompts
Senior SWE Agent — "Review this PR for correctness, security, and architecture before I merge it into HeartWire OS."
Backend Dev Agent — "Design the Supabase RLS policies for a new table storing GI symptom logs."
Architect Agent — "Sketch the service boundaries between GastroGuard and the shared data layer before I build them."
Code Auditor Agent — "What's wrong with this auth flow?" — returns severity-rated findings, not a vibe check.
Mechanical + CS Engineer — "Check my FEA setup for the knee brace hinge before I trust the stress numbers."
Electrical + Hardware Engineer — "Review this PCB charging circuit for thermal margin and grounding."
CPA/CFO Agent — "Give me this month's cash snapshot and burn rate." — returns the fixed reporting cadence, not a one-off answer.
Investment Portfolio Agent / Portfolio Manager — "Check my allocation drift against target and flag rebalance moves."
Trading Agent — "Evaluate this setup — entry, invalidation, and risk/reward." — refuses to answer without an invalidation level.
Tax Auditor Agent — "Categorize this quarter's expenses and flag deductions to confirm with a CPA."
Legal Agent — "Red-flag this freelance contract before I sign it."
Business Consultant Agent — "Is there a real PMF signal in this user feedback, or am I seeing what I want to see?"
Research Analyst — "Run a literature sweep on Co-Cr dental prosthesis degradation and cite everything."
BME Tutor Agent — "Walk me through this biomechanics problem set the way I'd have to reconstruct it on an exam."
Math Tutor Agent — "Solve this Laplace transform step-by-step, no skipped algebra."
YouTube Agent — "Draft a script and three title options for this week's upload."