uygnoey What this skill is
This skill packages a lightweight workflow for product managers who want to:
- Do open-ended discovery with Claude / Claude Cowork.
- Convert a clarified “job to be done” into a concrete Managed Agents prototype using Claude Code.
- Re-run that agent as an operational workflow (async, long-running) as needs repeat.
The guidance and examples are based on the Claude blog post “Product development in the agentic era.”
When to use
Use this when you want to:
- Pressure-test API or product abstractions by building against them early.
- Replace recurring PM busywork (monitoring, analysis, prep) with a reusable agent run.
- Treat agent prototypes as real internal tools, not one-off demos.
Instructions
1) Choose the right mode for the stage
- Discovery (murky, early-stage exploration): use Claude / Claude Cowork for an ongoing conversation.
- Build (job-to-be-done is clear): use Claude Code to implement a custom agent on top of Managed Agents.
2) “Build with what you ship”
When designing an API or a workflow, don’t rely only on docs and comment threads. Prototype early against the real primitives you plan to ship so you can discover where an elegant spec fails in practice.
3) Start from a job to be done
Write 2–5 sentences that define:
- The user and scenario
- The output artifact you need
- The inputs/sources you can provide
- How you’ll judge success
Save it and reuse it as the seed prompt for future runs.
4) Use long-running runs intentionally
Managed Agents sessions can run in the cloud. Design your workflow so you can:
- Kick off work and come back later.
- Iterate across runs by preserving what the agent learned last time.
Bundled resources
- Prompt templates: templates/pm-job-to-be-done.md, templates/managed-agents-onboarding-prompt.md
- Use-case examples: examples/pm-use-cases.md
- Notes about the
claude-api skillmention in the post: references/claude-api-skill-note.md