Harden Ponytail correctness guardrails

[ScottMarquis-PEC]

Summary

• keep explicit requirements and acceptance criteria mandatory in every intensity
• scope terse code-first output to implementation tasks while preserving full review/report evidence
• fix bugs at the narrowest layer that owns the violated invariant, not merely a shared downstream interception point
• require project-native verification, relevant existing checks, and safety-invariant coverage
• protect correctness, recovery, resource-cleanup, security, accessibility, and trust-boundary behavior from shortest-diff pressure
• keep hook fallback instructions and generated OpenClaw skill synchronized with the canonical skill

Why

A bounded four-lens review found that several minimalism rules could conflict with required scope, root-cause ownership, non-implementation outputs, and executable verification. This keeps Ponytail minimal without treating incomplete work as lazy.

Validation

• 68 root tests passed excluding the environment-dependent correctness probe
• 23 Pi extension tests passed
• 3 Ponytail MCP tests passed
• canonical skill, fallback instructions, and stale-wording assertions passed
• OpenClaw generated-copy parity passed
• git diff --check passed

Full npm test reached 81/82. The unchanged CSV correctness fixture fails locally because C:\Python314\python.exe has no pandas; no dependency was installed for this documentation/instruction change.