Run parallel subagents on one work unit and reconcile their results.
npx skills add https://github.com/pnewsam/skills --skill fan-out-pnewsam
Please help me install this Agent Skill. Skill: fan-out Source: https://github.com/pnewsam/skills/tree/main/registry/fan-out Command: npx skills add https://github.com/pnewsam/skills --skill fan-out-pnewsam
High-intent search queries and answers about installing and using this skill.
Use interrogate mode: send the same diff, intent, and context to several independent reviewers, each blind to the others, then reconcile findings by root cause. Include at least one different-lab frontier model and disclose the models used.
Arena runs competing attempts at the same brief and grafts the best ideas into one candidate. Swarm partitions a broad surface into slices for parallel coverage. Interrogate has several independent reviewers examine the same artifact and reconciles their findings.
Skip fan-out for routine, well-understood changes a single pass handles, since a frontier model on a small diff is already at ceiling. Fan-out pays off with scale: large surfaces, complex or high-consequence PRs, and wide design spaces.
The runbook requires a subagent or workflow substrate with parallel workers and result aggregation. Without one, run the closest sequential approximation and state the limitation rather than claiming a parallel result.
A frontier model from a different lab brings blind spots the primary model does not share, while a cheaper same-family model adds false positives without new signal. Findings raised independently by two models carry higher confidence.
zhixuan312Dispatch ad-hoc tasks to workers and return structured per-task reports.
tomzxAggregate and rank audit findings from multiple SDLC checks into a report.
microsoftCompare RPI planning and implementation evidence and record severity-graded review findings.
telefrekImplement small code changes with TDD and minimal setup.
