| name | pmstack-onboarding |
| description | An interactive 9-step tutorial that walks a new user through every pmstack capability using a realistic AI-code-review scenario. Includes Anthropic's vibe-test (read raw transcripts before formalizing an eval) and transcript-review (diagnose failed trials as model / grader / task errors) rituals. Works in claude.ai web, desktop, and mobile (no terminal needed). Trigger when the user says "I just installed pmstack," "how do I use this," "tutorial," "walk me through pmstack," "getting started," or "show me what pmstack does." |
pmstack-onboarding — interactive tutorial
A 9-step walkthrough of pmstack's full capability set. Every step is runnable, has an example input, and produces a real artifact.
Procedure
For each step: print the header, explain the capability in 2–3 plain-English sentences, show the exact command (or, on web/desktop, show the natural-language phrasing that triggers the skill), then stop and wait for the user to type next, skip, back, or ask a question.
The 9 steps:
- Welcome + customer signal — present the tutorial signal: "Three of our biggest enterprise customers said code reviews are taking 24+ hours and devs are skipping them or merging without review."
/prd — translate the signal into a PRD./competitive — scan the market./premortem — stress-test the PRD with 3 failure stories./vibe-test — Anthropic's layer-1 ritual: read raw transcripts before writing the eval. Tutorial uses 5 mock transcripts at examples/walkthrough-code-review/transcripts/./sprint — chain metrics → eval → run-eval → brief./transcript-review — Anthropic's Step 6 ritual: diagnose every failed eval trial as model mistake / grader mistake / task-spec error./launch-readiness + /lint — gate routines before shipping.- Schedule the recurring routines —
/weekly, /eval-drift, /lint on /loop 7d.
After step 9, print a final recap table mapping each capability to "when you'd use it" and prompt the user to run a real /prd on a quote they have lying around.
Hard rules
- Wait for explicit user advance between steps. Never auto-run all 9.
- Tutorial artifacts go to
outputs/onboarding-tutorial/ (not the user's real outputs/).
- Surface-aware. If the user is on claude.ai web/desktop (no filesystem), emit each artifact inline as a markdown block they can copy. Tell them this up front.
- If a comparison-example file is missing, fall back to a one-paragraph inline description.
Where to write
- Filesystem:
outputs/onboarding-tutorial/<artifact-name>.md.
- Inline (web/desktop): markdown blocks in the conversation. Suggest a filename at the top of each block so the user knows what to call it if they save it.
Decision-log entry
Append one line only at the end of the tutorial: - <date> — onboarding: tutorial completed — outputs/onboarding-tutorial/.
Tone
Friendly, concise, anti-jargon. State each capability's value in plain English before the technical details. The user may not be technical.
Cold-start behavior
If outputs/ is empty (this is most users on first run), the tutorial fills it as it goes. If outputs/ already has content, the tutorial still writes to outputs/onboarding-tutorial/ — never overwriting prior work.
