01 · shape
Claude for ambiguity
Product calls, architecture, confusing requirements, and the hardest review pass.
free developer workshop
Dated rankings, built for real work.
Models, coding surfaces, agent skills, and MCP servers ranked for Waterfall fit, not universal supremacy. Every pick says what it is for, why it earned a place, and when to leave it off.
Claude plus Codex
Use both every week.
Do not burn one plan to zero and then drag a cold project into the other. Give each agent a distinct job and preserve a compact shared handoff.
02 · build
Codex for implementation
Repository work, parallel agents, worktrees, tests, and autonomous execution.
03 · challenge
Cross-review the result
One agent owns the change. The other gets a narrow review brief and no incentive to defend it.
Loading the workshop catalog...
How picks earn a place
Four ranking gates.
A popular tool can still lose if it creates more context, permission, billing, or maintenance cost than useful work. skills.sh installs and mcp.directory discovery are signals, not quality or security verdicts.
01 · task fit
A named recurring job, not a generic promise to make agents better.
02 · evidence
Official sources, dated benchmarks, or observed local use.
03 · negative case
An honest reason to keep it off or choose something else.
04 · safe source
No copied secrets, machine config, or unlicensed third-party code.