Skip to content

init · guide · run · version

Install the PDSA skill into a workspace so an agent can trigger the loop.

Terminal window
pdsa init # writes .claude/skills/pdsa/SKILL.md
pdsa init --lang en # or --lang ko

See For AI Agents.

One-off PDSA advice from the LLM (a simple pass-through prompt — it does not record a cycle).

Terminal window
pdsa guide "<question / situation>"

Run the demo PDSA feedback cycle — the reference PdsaLoop implemented as a real Akka.Streams cycle — and record the run to the graph DB. This is the learning demo, distinct from the agent workflow commands.

Terminal window
pdsa run

Check the latest version on the npm registry and update the global install.

Terminal window
pdsa update # check latest, then update (npm global)
pdsa update --check # only report current vs. latest (no install)
  • The pdsa help / no-argument screen also shows a “new version available” note when you’re behind (24 h cached, so it stays instant and works offline).
  • Pre-check, no force-kill: if another pdsa process is running, update does not kill it. On Windows a running instance locks the native binary, so update asks you to close it first, then retries.
  • Self-lock avoidance: on Windows the actual npm i -g runs in a fresh console while pdsa exits, so the running executable/DLL is unlocked before replacement; on Linux/macOS it runs inline.

Print the version, .NET runtime, and stack.

Terminal window
pdsa version
# pdsa <version>
# .NET <runtime>
# Stack: Akka.Streams (PDSA feedback cycle) · Kùzu embedded graph DB · OpenAI