Use flow repro evidence instead of failing-test-first guidance

[luca-schweigmann]

Fixes #228.

Bug-fix step 5 now keeps before/after repro evidence on the real surface or in the flow's existing deterministic E2E, without a new unit test or failing-test commit. Refactoring step 1 reuses that flow's E2E and adds its one deterministic E2E only when uncovered. The sequencing principle uses repro/baseline-then-fix; session-start drops direct TDD entry. The type-checked build remains a build step. CHANGES.md and UPSTREAM.md record the local adaptation. No new tests.

Assumptions: trunk is current ericlitman/open-pstack main per UPSTREAM.md, rather than the local luca branch. This PR changes only the four named instruction sites plus provenance; remaining contradictory guidance is intentionally unchanged. No candidate install, live-gate publication, or merge is part of this task.

Checks

• bun run test: passed (Node suite and Bun runner suite; Bun: 502 passed, 0 failed).
• bun run typecheck: passed.
• bun run static: passed (includes manifest/static invariants).
• claude plugin validate ./plugins/pstack: passed.
• git diff --check: passed.

Follow-up candidates (grep only; unchanged)

Paths below are relative to plugins/pstack/.

• skills/tdd/SKILL.md: cheap-local-test auto-trigger, unit/component/integration preference, new failing regression test, and sibling coverage.
• skills/principle-test-behavior-not-implementation/SKILL.md: guidance to write/rewrite tests, mocked collaborator cases, and compile-time invariant tests; does not restrict the mechanism to flow E2E.
• skills/principle-foundational-thinking/SKILL.md: “tests before fixes” and test infrastructure as scaffold.
• skills/blast-radius/SKILL.md: “Write a script or test”.
• skills/poteto-help/SKILL.md and skills/poteto-help/references/recipes.md: cheap-test bug routes to TDD.
• skills/poteto-mode/playbooks/autopilot-full.md: fix-forward requires a red test across defect sites.
• skills/poteto-mode/playbooks/orchestrate.md: frame and ledger accept unit-test-verified as a behavioral verdict.
• skills/poteto-mode/scripts/orch/store.ts: matching unit-test-verified verdict type, ranking and validation wording.

These are candidates for a separate scoped change, not claims that every use of “test” requires removal.

[mergify]

Merge Protections

🟠 1 of 2 protections blocking · waiting on 🤖 CI

	Protection	Waiting on
🟠	Queue ready pull requests	🤖 CI
🟢	🚦 Auto-queue	—

🟠 Queue ready pull requests

Waiting for

• check-success=Unfret
• check-success=live-gate
• check-success=verify

Waiting checks: Unfret, live-gate, verify.

Automatically queue non-draft pull requests targeting main after verify, Unfret, and live-gate pass on the head and every review thread is resolved.

• check-success=Unfret
• check-success=live-gate
• check-success=verify
• #review-threads-unresolved = 0
• -draft

---

Show 1 satisfied protection

🟢 🚦 Auto-queue

When all merge protections are satisfied and these conditions match, this pull request will be queued automatically.

• -files~=^\.github/
• base = main

[greptile-apps]

[Medium impact] The instruction changes appear safe to merge once the required exact-head live gate passes.

Summary

Replaces failing-test-first guidance at four instruction sites with before-and-after proof from the affected flow.

• Bug fixes use repro evidence instead of test-first commits.
• Refactors check the flow’s behavior with its deterministic E2E.

No actionable defects were established. This review did not rerun the reported checks or perform a live installation.

luca-schweigmann explicitly deferred the remaining contradictory test guidance to a separate scoped change. luca-schweigmann also excluded candidate installation and live-gate publication from this task; AGENTS.md permits review before the live test.

_{Reviews (1) · Last reviewed commit: "Use flow repro evidence instead of faili..." · Reviewed by Greptile}

[luca-schweigmann]

Closing: this was a local experiment for our fork; no upstream change intended.