pstack: parity sweep with the private skill tree

[poteto]

Summary

Audit of pstack against its private upstream skill tree found drift in the playbooks and two defects (a wrong model-count label and a dangling skill reference). This PR ports the missing content in pstack's public naming and reference style, keeps pstack's existing model defaults, and fixes both defects.

Changes

1. hillclimb.md: ported the upstream steps 1-2. Step 1 now grounds the workload and architecture before choosing the ruler (run how first, name workload dimensions, require a case that reproduces the complaint). Step 2 now proves the harness's sensitivity with contrasting workloads before freezing it. Step 4 rewritten to reference the step-1 architecture model instead of a redundant up-front how run.
2. refactoring.md: inserted the missing step 2, naming the structure the code is missing per principle-model-the-domain (state machine over scattered booleans, table/registry over branching, typed model over shape assumptions, reducer over ad hoc mutations), and renumbered steps 3-8. Added the intro framing sentence "A refactor that smuggles in a behavior change loses its safety net."
3. feature.md + poteto-mode SKILL.md: the delegation scope now names the data shape and its organizing structure per principle-model-the-domain, chosen before the delegate writes logic. The "Any code" trigger gets the same expansion.
4. typescript-best-practices/SKILL.md: restored two dropped rules in generic form. "Real tests" (don't mock what you can run; leak/disposable checks; verify UI in a running build) and "Structured telemetry" (structured logger diagnostics, no console.log in shipped code).
5. poteto-mode Subagents: added the tiering criteria. Hardest changes go to the strongest judgment model when the task needs judgment or intent is vague, to the strongest instruction-following model when the work is a precisely specified sequence, and trivial mechanical edits to the fast code model. Existing pstack slugs kept. Also added the override semantics: per-role lines in the /setup-pstack rule override these defaults and the routed skills' choices; a role with no line keeps its default.
6. One-liner tells, verbatim: bug-fix step 1 ("A bug you can't reproduce, you can't prove fixed."), bug-fix step 2 and runtime-forensics step 3 (the plausible-but-unconfirmed cause rationales), session-pickup step 3 (the "verify from scratch" tell), autonomous-run steps 1 and 4 ("A vague goal stalls; a predicate lets you stop." / "A run with no trail can't be audited or resumed."), authoring-a-skill ("Tell it to do the thing and skip the reason. Explain only when the rule is confusing without one.").
7. Bug fix: poteto-mode's interrogate trigger said "(four-model adversarial)" but the default panel is three models. Now "(multi-model adversarial)".
8. Leak fix: why/references/sources/databricks.md cited a databricks-use-dbt-models skill that doesn't exist publicly. The sentence now stands alone (confirm model names with SHOW TABLES).

Plus: arena cross-judge pool role line added to setup-pstack's rule shape (Arena picks one judge from it on a different family from the parent's), and plugin version bumped 0.11.2 -> 0.11.3.

Verification

• Leak check over every touched file (rg -i 'lauren|everysphere|anysphere|cervino|rainier|kindle|databricks-use'): zero hits in content (the only match is the pre-existing author field in plugin.json).
• refactoring.md steps verified sequential 1-8; cross-references (bug-fix "step-3", SKILL.md "Feature step 3", hillclimb "step 1") verified intact.
• No internal model slugs, tools, or channels introduced; pstack's existing defaults unchanged.

---

Note

Low Risk
Documentation-only changes to Cursor plugin skills and playbooks; no runtime application code or security-sensitive logic.

Overview
Brings the public pstack skill tree in line with a private upstream audit: mostly agent instruction updates in poteto-mode playbooks and related skills, plus a patch version bump.

Workflow rigor. Hillclimb now grounds workload/architecture and proves the measurement harness before freezing it. Refactoring gains an explicit model-the-domain step (name the missing structure) and renumbered follow-on steps. Feature and the Any code trigger require naming organizing structure, not only the data shape.

Delegation and config. Subagents documents difficulty-based model tiering and how /setup-pstack overrides routed-skill defaults. setup-pstack adds an arena cross-judge pool line and documents how Arena picks a cross-family judge.

Fixes and hygiene. Interrogate trigger text changes from inaccurate four-model to multi-model. databricks.md drops a reference to a non-public skill. typescript-best-practices restores Real tests and Structured telemetry rules. Several playbooks gain short “tell” lines (repro, unconfirmed cause, audit trail, skill voice). Plugin version 0.11.2 → 0.11.3.

^{Reviewed by Cursor Bugbot for commit b37d905. Bugbot is set up for automated code reviews on this repo. Configure here.}