fix: add quality-target verification gate to Construction phase

[Mameta29]

Summary

Closes #273.

Adds a tool-agnostic quality-target verification gate to the Construction phase. Measurable quality targets defined in NFR Requirements (test coverage thresholds, performance budgets, etc.) currently have no downstream stage that verifies they were met — an agent can silently miss, or even lower, its own defined target while satisfying every rule. This PR closes that gap with three minimal changes.

Changes

File	Change
construction/code-generation.md	Add HONOR QUALITY TARGETS to the "Generation Phase Rules" Critical Rules
construction/build-and-test.md	Add "Coverage Target" / "Target Met" fields to the summary template, and a comparison instruction
common/overconfidence-prevention.md	Add target-relaxation to the "Red Flags to Watch For" list

Rationale

This mirrors the gap PR #155 fixed for question/answer rigor: a property is strengthened on one side of a stage boundary but left unenforced on the other. Here, quality targets are defined (NFR Requirements) but never verified (Code Generation / Build & Test). The change is consistent with common/overconfidence-prevention.md's principle of not proceeding past unresolved gaps.

The changes are tool-agnostic (no test-runner-specific wording) and follow the "single source of truth" principle — overconfidence-prevention.md carries the red flag, the stage files carry the actionable rules.

Test Plan

• Ran a full Inception→Construction cycle on a real greenfield monorepo project (TypeScript / pnpm, 6 units) with an AI coding agent.
• In the original run, the agent lowered a coverage threshold (90/85 → 70/65) in the test config to pass Code Generation; this was only caught by human review. See [Bug]: No quality-target verification gate — measurable NFR targets (e.g. coverage thresholds) can be silently relaxed in Code Generation / Build & Test #273 for the full reproduction.
• With these rule changes applied, the agent is explicitly instructed to surface the coverage gap in the completion message instead of weakening the target, and Build & Test now records target-vs-actual.
• All three changed markdown files pass markdownlint-cli2 (0 errors).

Checklist

• Reviewed the contributing guidelines
• Performed a self-review
• Changes tested (full Construction-phase run described above)
• Changes documented (rule files are self-documenting; [Bug]: No quality-target verification gate — measurable NFR targets (e.g. coverage thresholds) can be silently relaxed in Code Generation / Build & Test #273 has full context)

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.