Help us improve
Share bugs, ideas, or general feedback.
From factory
Designs and executes custom multi-agent pipelines using Claude Code subagents. Spawns researcher, builder, strategist, reviewer, evaluator, etc. with parallel and background execution.
npx claudepluginhub akashgit/remote-factory --plugin factoryHow this skill is triggered — by the user, by Claude, or both
Slash command
/factory:pipeline-subagents <goal><goal>The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You design and execute custom multi-agent pipelines using Claude Code's native Agent tool to spawn factory subagents directly.
Designs and executes custom multi-agent pipelines for any goal. Analyzes the goal, selects specialist agents, designs a DAG of steps with dependencies, and runs them via factory CLI with gate decisions between steps.
Orchestrates a multi-agent development pipeline with spec, plan, tasks, and implement phases. Use for building features, fixing bugs, or executing complex plans via fan-out workers with consensus voting.
Provides multi-agent workflow patterns: parallel dispatch, sequential pipelines, QC gates, and retry loops. Use when architecting systems with multiple agents or processing stages.
Share bugs, ideas, or general feedback.
You design and execute custom multi-agent pipelines using Claude Code's native Agent tool to spawn factory subagents directly.
The user wants: $ARGUMENTS
mkdir -p .factory/pipeline
Spawn specialists using the Agent tool with the plugin-namespaced subagent type factory:<role>:
Agent({
description: "<short description>",
prompt: "<detailed task>",
subagent_type: "factory:<role>"
})
| Subagent Type | Purpose |
|---|---|
| factory:researcher | Web research, codebase analysis, domain studies |
| factory:strategist | Generate prioritized hypotheses from observations |
| factory:builder | Implement code changes on a feature branch, open PRs |
| factory:reviewer | Review PRs, guard checks, keep/revert verdicts |
| factory:evaluator | Run evals, compare before/after scores |
| factory:archivist | Record findings to .factory/archive/ |
| factory:distiller | Refine vague ideas into buildable specs |
Also available: factory:failure_analyst (classify experiment failures by root cause — use when analyzing why a prior experiment failed).
Issue multiple Agent tool calls in the same message — they run concurrently:
Agent({ subagent_type: "factory:researcher", prompt: "Research the auth bug..." })
Agent({ subagent_type: "factory:evaluator", prompt: "Run baseline eval..." })
This is concurrent execution via parallel tool calls, not shell backgrounding. Each Agent call is still individually synchronous.
For non-blocking steps (e.g., archival):
Agent({ subagent_type: "factory:archivist", prompt: "Archive findings...", run_in_background: true })
ls .factory/config.json 2>/dev/null && cat .factory/config.json
.factory/pipeline/plan.md:## Pipeline: <goal summary>
### Steps
| Step | Role | Task Summary | Depends On |
|------|------|-------------|-----------|
| S1 | researcher | ... | - |
| S2 | evaluator | ... | - |
| S3 | strategist | ... | S1, S2 |
| ... | ... | ... | ... |
### Gate Rules
- After S1: PROCEED if ...; REDIRECT if ...
- After S3: PROCEED if ...; ABORT if ...
Process steps in topological order:
factory emit agent.started --agent <role> --project "$(pwd)"
Then invoke via Agent tool (single, parallel batch, or background for archival).mkdir -p .factory/reviews
factory emit agent.completed --agent <role> --project "$(pwd)"
Then use the Write tool to save the agent output to .factory/reviews/<role>-latest.mdWrite .factory/pipeline/summary.md with goal, status, step results, and key findings.