Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open

Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open source project that replaced Claude Design: Fireworks-design currently only supports ClaudeCode. If it exceeds 2k stars, it will also be considered to be compatible with Codex. ps: Use the latest GLM 5.2 as the main coding model. A single model generation is essentially a draw from the distribution - its taste, current status, and understanding of needs are all locked in that version, and the upper limit of quality is stuck by the variance. So I no longer bet on "whether the draw is good this time", but deliberately draw more times and let a group of independent judges eliminate the losers. This is also the sentence that struck me in the ClaudeDesign announcement: Even senior designers can only "explore with restraint" and rarely have time to pursue more than a dozen directions. Then leave it to the assembly line - 8 completely different aesthetic directions are rolled out at one time, 6 dimensions are independently scored, the champion makes the skeleton, and the rest contribute highlights, and then the review↔repair cycle is completed until it passes the line. Different temperament needs will choose different winners (Dark Premium is chosen for the cinematic texture of the movie category, and Bold Editorial is chosen for most technical tools). This is the value of a review, not a final decision. Technically, it is not a "generate it once and then let the model change itself" cycle. It is a Claude Code Dynamic Workflow - multi-agent orchestration, which welds together the most practical things in the current reasoning period: best-of-N + self-consistency, LLM-as-judge review panel, critique-and-revise (Self-Refine), and structured output of schema verification. The control flow belongs to the script, not the model, so it is reproducible, recoverable and model-independent. This is also the latest Agent direction that has been recommended recently.

Visual summary

Article argument map

Generated from the post's content graph

Mermaid outline

flowchart LR
  format-long_post["longform note"]
  topic-agents["agents"]
  topic-technical-distribution["technical distribution"]
  topic-tooling["tooling"]
  capability-agent-workflow["agent workflow"]
  capability-evaluation["evaluation"]
  capability-technical-distribution["technical distribution"]
  market-open-source-builders["open-source builders"]
  format-long_post -->|covers| topic-agents
  format-long_post -->|covers| topic-technical-distribution
  format-long_post -->|covers| topic-tooling
  format-long_post -->|signals| capability-agent-workflow
  format-long_post -->|signals| capability-evaluation
  format-long_post -->|signals| capability-technical-distribution
  format-long_post -->|signals| market-open-source-builders

Visual structure

Essay structure map

Built from summary and key paragraph positions

Mermaid outline

flowchart LR
  thesis["Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open so..."]
  signal["Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open so..."]
  operator["Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open so..."]
  implication["Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open so..."]
  thesis -->|frames| signal
  signal -->|develops| operator
  operator -->|lands in| implication

Source: View the original post