Tips for getting the best output from the parallel agents

[Rishiidev]

The agent runs 16 agents across 3 phases. The quality of the output depends a lot on what you feed it at Step 0.

Tips that improve output quality

On the value prop question:
Include a number if you have one. The agent uses this for the README headline and the social preview. "Cuts launch time by 75%" lands better than "saves time."

On the repo name:
Let the agent score names rather than giving it a finished name. The name scoring phase (Step -1) runs 5 candidates through SEO and memorability criteria. If you hand it the name you already decided, you skip this.

On the .skill file:
If you have one, point the agent at it early. Agent A uses it for trigger phrase extraction and platform detection. Without it, those fields are inferred and occasionally wrong.

On competitor research (Agent B):
The more specific your project description, the better the competitor search results. "A Claude skill that launches GitHub repos" finds better comparisons than "a developer tool."

Common gotchas

Issue	Fix
Agent C generates SVG but it does not embed	Make sure the README file exists before Phase 1 runs — create a stub if needed
Trigger phrases not activating	Check that the .claude-plugin folder is at the repo root, not in a subfolder
GitHub Pages not enabling	Confirm your account has Pages enabled and the repo is public
Distribution content feels generic	Add more specific context about your target user in Step 0

What tips do you have? Drop them below.