← @bgSponge

The Complete Guide to pstack Pt. 2

Summary

Lauren (@poteto) argues that effective agentic engineering requires replacing abstract planning with code-based prototyping and rigorous verification using the pstack workflow.

  1. Context priming prevents agent failure.

    Agents fail when they lack intent specification or sufficient context; high-quality context priming is required to ensure superior outcomes.

    Using agents well comes down to how well you're able to prime the agent's context window with high quality context.

  2. Indirect prompting clarifies intent.

    Asking agents to restate problems in their own words prevents the user from imposing incorrect assumptions or leading the agent astray.

    Instead of telling the agent exactly what I want, I try to draw it out of the agent

  3. The /teach skill builds trust.

    The /teach skill uses /how to trace mechanics and /why to query historical evidence like Git, Linear, and Slack to explain reasoning.

    When you run /why, pstack queries historical evidence across multiple sources in parallel

  4. Prototyping replaces abstract planning.

    The prototyping playbook uses throwaway sketches and /control-app to provide empirical evidence instead of relying on theoretical, over-specified plans.

    Prototyping is planning, but with code.

  5. Architecting requires parallel competition.

    The /architect skill spawns independent candidate runners across different model families to sketch designs, which a cross-judge agent then evaluates.

    It spawns independent candidate runners in parallel, often across different model families.

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →