Turning an Agent into a Chief Architect
shazanwhite · reddit · 2026-07-14
The author open-sourced an Agent Skill based on the agentskills.io / skills.sh format. The goal isn't just "letting AI write code," but forcing coding agents to make decisions like a Chief Software Architect.
Core practices include: defining goals, constraints, and non-goals upfront; introducing reverse costs when comparing solutions; explaining "what not to optimize" alongside the recommended path; applying a PRIMARY → ADVERSARY → SYNTH workflow for larger reviews; and generating baselines and tiered issue lists with stable IDs. It supports Markdown + frontmatter and uses Claude Code / Grok / Cursor / Codex as harnesses, installed via skills.sh.
The author's key takeaways:
- A good skill is a "workflow + knowledge directory + anti-patterns," not just a prompt saying "act like an architect."
- Stable machine tokens + natural user language descriptions are more effective.
- Unit-level baselines reduce vague reviews.
- Adversarial reviews mitigate overconfidence.
- Blank baselines are merely compliance theater.
Related event: Open-Source Project Elevates AI Coding Agents to Software Architects(3 posts)→
More from coding & agent
- A better path to agent autonomy is running waves, finding friction, and iterating — JnBrymn · 2026-07-22
- Coding agents are heading toward an AI-writes, AI-reviews, human-approves workflow — aftahi_ai · 2026-07-22
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- GitHub review bot hits its PR limit and forces a 39-minute cooldown — DanielLockyer · 2026-07-22
- Max reasoning effort appears to be mobile-only in Codex Remote, not desktop — GabGarrett · 2026-07-22
- A Reddit demo argues online stores should expose carts and pricing through MCP — gelembjuk · 2026-07-22