feat(pstack): port explain-the-number, fresh subagents, hourly autopilot tick, PR headings, schema-first cast (0.15.6)

[poteto]

Why

pstack drifted behind the source mode skill on five changes. This port brings them over in genericized form.

What changed

A. Explain the number. New principle-explain-the-number and benchmark-checklist skills. Adds the trigger and index lines in poteto-mode, perf-issue step 1 and hillclimb step 2 now vet numbers with the checklist, and the README rows and guide count go from 23 to 24.

B. Fresh subagents by default. Adds a poteto-mode paragraph about spawning a fresh subagent for each task. The AskQuestion trigger now says to explain each alternative in plain words. Updates the poteto-agent description, and swarm Phase C now says "respawn" instead of "rerun".

C. Hourly autopilot tick. autopilot-full, autopilot-stack, and multi-phase-plan drop /goal and move the tick from 30 minutes to /loop 1h. The owner pushes after every verifiable unit, the tick checks unrelated trunk moves, and the operator judges the owner by the pushed branch and the decisions trail. check-plan.mjs drops the /goal marker and the 30-minute pattern and adds /loop 1h.

D. PR body headings. opening-a-pr.md now uses ## Why / What changed / Scope / Tradeoffs / Blast Radius / Verification and adds concision rules. The built-in PR tool rule is merged with the existing draft: false note, and the forge section (gh by default, origin when available) stays. One line in multi-phase-plan now follows the same rule.

E. Schema-first cast. typescript-best-practices/references/patterns.md replaces the hand-validated parseUser "earned cast" with a Zod schema plus z.infer, and adds the z.ZodType<User> annotation rule for a type declared first.

Version bump: 0.15.5 to 0.15.6.

Scope

18 files under pstack/. Unslop, model routing, and the audit's "needs a decision" items are not touched.

Tradeoffs

• The cloud-run wake guidance is removed. Cursor's public cloud agent docs (cursor.com/docs/cloud-agent/capabilities) say that cloud agents support timers and /loop, so one /loop 1h instruction covers local and cloud runs.
• Overlap with open contributor PRs:
  • fix(pstack): make the /poteto-mode chat the runner, poteto-agent a delegate only #358 goes the opposite way from B. It makes the /poteto-mode chat the runner and poteto-agent a delegate only, and it edits the same poteto-agent.md and poteto-mode/SKILL.md. Only one of the two directions can land.
  • fix(pstack): kebab-case agent slugs with no spaces #378 also edits the poteto-agent.md frontmatter and the README.
  • feat(pstack): bundle deslop skill #371 also edits opening-a-pr.md (D), poteto-mode, and the README.
  • pstack: teach explicit variable naming in minimize-reader-load #232 also edits 08-principles.md and the README principles table (A).
  • fix(pstack): handle tiny known bug fixes inline #363 and poteto-mode: fetch latest main before starting work or spawning agents #377 also edit poteto-mode/SKILL.md.
  • These are line-level conflicts. The PR that merges second needs a small rebase, and if it also bumps plugin.json (fix(pstack): make the /poteto-mode chat the runner, poteto-agent a delegate only #358 does), it needs a new version number.

Blast Radius

Text-only skill and playbook changes plus one marker list in check-plan.mjs. Plans written against the old checker that contain /goal or a 30-minute tick now fail the check until they say /loop 1h.

Verification

• node scripts/validate-plugins.mjs with ajv and ajv-formats: All plugins validated successfully.
• check-plan.mjs on a filled plan from the new template: 0 problems, exit 0.
• Zod example compiled with tsc --strict. The schema-first file passes, and a mismatched z.ZodType<User> fails with TS2322.
• Leak sweep (grep -i -E) over every added line of the diff: 0 hits. Over the whole pstack/ tree, no hits beyond the hits that main already has (the LICENSE and manifest author, the standard .cursor/skills/ paths, the published benny automation, a "magnifying glass" image alt text, a glint-482 example in a script comment, and the existing public model names). Patterns:
  \blauren\b lauren-mode lauren-agent \bTan\b \bDylan\b dgattey \.cursor/skills/ check-glass-perf-regression \bGlass\b anytool /in-cloud \bbcId\b dev check \bsand\b \bSand\b \bbenny\b \bsandy\b fetch_cloud_agent_debug_info cursor-subscriptions subscribe_timer cursor\.com/codebase cursor\.com/canvas cursor\.com/agents slack\.com linear\.app GLINT- everysphere anysphere origin\.cursor\.com opus-5-5-xhigh \bFable\b rainier Tahoma GPT-5\.6 \bSol\b ablation sign test p = 0\. \.buildkite

  

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 790b45c. Configure here.}