docs(pstack): refresh guide for /correct, checklist, prompting tips

[poteto]

Brings pstack/docs/guide/ up to date with the skills that ship today and with the durable prompting habits from the published pstack articles and tip posts. The guide voice stays the same: no attribution, no X links, and habits and copyable prompts instead of version history. Only files under pstack/ change. Version goes from 0.15.12 to 0.15.13.

Skill gaps closed

Skill	Where	What the guide now says
/correct	09 (new section), 07, 08, 10	The fix-the-environment ranking (architecture, then types and lints, then tests, docs last). How it finds mistake classes, proves each check against a real past mistake, and keeps the rule table.
/benchmark-checklist	06 (new section), 05, 08, 10	The seven questions, the four verdicts, when to type it yourself, and that the Perf issue and Hillclimb playbooks already run it. Linked from Explain the Number.
/poteto-help	README, 10	You type it with a question. It answers, gives you a prompt, links the source, and does not start the work.
/make-bot-ui	09	A short section on this situational Grok Bot skill.

Habits folded in, by page

• 02 Poteto mode: what goes in a prompt and what to leave out, restating a noisy report before any code, short follow-ups, and cloud subagents (/in-cloud) as the first choice for parallel work, with worktrees as the local fallback.
• 03 Understand: supervising an agent that may know the code better than you. Covers the read-only Investigation playbook, /teach on the agent's own trade-offs, and /recall before new input.
• 04 Design: never take the first design. Prototype instead of debating, write the README or tutorial first for shared code, /architect throws away a design that needs any or the same workaround in many places, no adversarial review of plans that have no code behind them, and the Multi-phase plan playbook after the design settles.
• 01 Setup: tells new users to accept /create-verification-skill, and adds a short section on keeping cost down.
• 05 Build: the Runtime forensics and Trace forensics playbooks.
• 06 Verify: proof as artifacts you can inspect, prompts that name the verify skill, treating the verify skill as shared infrastructure, traits of an agent-friendly control CLI (build the lever), daily /maintain-verification-skill, and the Visual parity playbook.
• 07 Overnight: four checks a loop must pass before it runs alone, the Pause safely playbook, running many Cursor Projects with poteto-mode (including a sample prompt that uses /correct and /architect), and loops that start themselves through the dormant benny automation pack.
• 08 Principles: Build the Lever gets the habit "have the agent write the tool it wishes it had". Explain the Number links to /benchmark-checklist, and Encode Lessons in Structure links to /correct.
• 09 Make it yours: start with few skills, the update-and-eval prompt for skill edits, /correct, and /make-bot-ui.
• 10 Recipes and pitfalls: seven new recipes and six new pitfalls that match the habits above.

Checks

• Relative-link check over pstack/docs/guide/, anchors included: 0 broken links.
• node scripts/validate-plugins.mjs: all plugins validated.
• Counts still match the repo: 24 principles, 23 playbooks.
• I checked each claim about a skill or playbook against its file: /correct, /benchmark-checklist, /architect (the scrap phase), Multi-phase plan, Pause safely, /poteto-help, /make-bot-ui, and the benny README.
• I checked the two external links: cursor.com/blog/projects and cursor.com/docs/subagents#cloud-subagents.

Do not merge yet. This PR is for review.