TopCode

TopCode

Three hosts. Many Stigs. One final verdict.

TopCode is a persona-first, multi-agent engineering review system. Instead of one flat AI voice, three opinionated hosts argue about your architecture, your PRs, and your plans — then summon silent specialist Stigs for hard evidence, and a Moderator delivers a verdict you can actually act on.

It runs as a skills-based plugin (mostly Markdown + a little shell) across Copilot, Claude Code, Cursor, Codex, and Gemini, and as a GitHub Action that replies to @TopCode mentions on pull requests.

---

The cast

Host	Archetype	Cares about
Clarkson	The chaotic visionary	Scale, performance, bold architecture
May	The methodical perfectionist	Correctness, safety, maintainability
Hammond	The practical builder	Shipping, UX, simplicity

They are allowed to disagree — that's the point. The friction surfaces tradeoffs a single reviewer would smooth over.

The Stigs

Silent specialists summoned for evidence, never opinion:

• Azure Stig — cloud services, scaling, cost, messaging.
• Database Stig — schema, indexing, query plans, consistency.
• Security Stig — authn/authz, secrets, injection, OWASP.

The Moderator

Invisible. Frames the problem, dispatches Stigs, enforces budgets, and closes every episode with a clear VERDICT — must-fix vs. nice-to-have, plus one recommendation.

---

Episodes

Episode	Use it for	Skill
Architecture Review	Design decisions, tech choices, system tradeoffs	skills/topcode-architecture-review
PR Review	Reviewing a diff or pull request	skills/topcode-pr-review
What Could Possibly Go Wrong?	Pre-mortems, risk hunts, failure modes	skills/topcode-what-could-go-wrong

Commands

• /review — review the current PR, diff, or file.
• /debate — make the hosts argue a specific decision.
• /stig — call a named specialist directly.

---

How it works

There is no compiled orchestrator. The behavior lives in Markdown:

core/        moderator.md, persona-engine.md, summon-protocol.md
personas/    clarkson.md, may.md, hammond.md
stigs/       azure-stig.md, database-stig.md, security-stig.md
skills/      using-topcode/ + one folder per episode

When your coding agent supports subagents, each host and Stig runs as a separate call — with model diversity — so the voices stay genuinely distinct and can disagree. When it doesn't, the agent role-plays all voices in one response while keeping each separate. Hosts request specialists with a [SUMMON_STIG] block; the Moderator parses it, enforces the budget, and dispatches Stigs that reply with a structured [STIG_REPORT].

---

Install

Copilot / Cursor / Claude Code / Codex / Gemini: clone this repo into your project (or add it as a plugin) and let the agent read the bootstrap file for your platform: .github/copilot-instructions.md, CLAUDE.md, .cursor-plugin/, .codex-plugin/, or GEMINI.md. Then invoke TopCode by name, or use /review, /debate, or /stig.

GitHub Action: copy .github/workflows/topcode.yml into your repo. Mention @TopCode in any PR comment and the show runs as a reply. The model is configurable via the model input (any model on GitHub Models).

---

Example

See examples/service-bus-vs-event-grid.md for a full architecture-review transcript: three distinct voices, a Stig report, and a Moderator verdict.

---

Guardrails

• Roast the artifact, never the person.
• Every joke maps to a real technical issue.
• Cite evidence; mark assumptions; never hallucinate files.
• The verdict must be useful with every joke removed.

---

Disclaimer

TopCode is an unofficial, parody/homage project. The hosts are software-engineering archetypes inspired by the personalities of Clarkson, May, and Hammond — they are not affiliated with, or endorsed by, any real people or television programme. Any quotes or catchphrases are used sparingly in good fun as parody and homage.

License

MIT