Milestone 18: Bootstrap doctor/audit contract test expansion

[Copilot]

Adds explicit contract tests for all key target-repo diagnosis semantics in bootstrap_doctor.py, which were previously untested despite being stable operator-facing contracts (Milestone 16). Proves classification stability and anti-drift for the doctor layer.

New: tests/test_bootstrap_doctor.py (77 tests across 12 test classes)

• All six health state classifications via classify_health() — including priority ordering (stale-version beats profile-mismatch), patch-only version differences not triggering stale state, and low-confidence suggestions not triggering profile mismatch
• Version comparison helpers — _semver_tuple() and _is_materially_behind() unit-tested directly
• Status helpers — marker_status(), required_files_status(), placeholder_status(), profile_alignment() (all 4 alignment states)
• recommend_actions() for every health state — including an explicit assertion that --force never appears in any recommendation
• Era classification alignment — proves bootstrap_doctor.classify_era() always agrees with bootstrap_core.classify_marker_era()
• audit() integration tests using real temporary directories for 5 key states; includes a read-only proof that audit() does not mutate the target directory
• Health state name stability contract — asserts HEALTH_LABELS and HEALTH_DESCRIPTIONS cover exactly the six documented states

Integration tests use a self-contained fake source root with a pinned VERSION file (_TEST_MARKER_VERSION = "0.14.0") so tests are version-independent and don't break on future version bumps.

Infrastructure updates

• tests/test_bootstrap_doctor.py added to BOOTSTRAP_REPO_REQUIRED_FILES in validate_bootstrap.py and bootstrap-manifest.yaml
• AGENTS.md — expanded "Bootstrap doctor" rules section with anti-drift contract: diagnostic semantics are test-backed, output phrasing can vary but semantic meaning must not, shared helpers (classify_era, parse_marker) require updates to both test files when changed
• README.md — updated layout tree and Contract tests section to document doctor test coverage and the conservative-behavior guarantees the tests prove
• VERSION bumped 0.14.0 → 0.15.0; CHANGELOG.md updated

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[incredincomp]

@copilot
verify before merging:

the tests are asserting semantic meaning, not overfitting to phrasing or incidental implementation details; the summary suggests that is already the intent.

CI/checks actually run cleanly once GitHub finishes loading them; the page showed some GitHub “Uh oh!” load errors in the conversation view, so I could read the Copilot summary but not rely on the page for a perfect end-to-end check-status view.

[Copilot]

@copilot
verify before merging:

...

Verified both items:

CI: clean. The CI run on the latest commit (b60511f) completed with success — py_compile, validate_bootstrap, 120 contract tests, and the fixture self-test all passed.

Tests: semantic meaning, not phrasing. A few specifics:

• TestClassifyHealth / TestClassifyEra / TestProfileAlignment / TestMarkerStatus etc. all assert on return-value strings from semantic functions (e.g., "populated-and-healthy", "aligned", "pre-version"), not on human-readable label or description text.
• TestRecommendActions asserts which tool appears in the recommendations (suggest_profile.py, refresh_bootstrap.py, etc.) and which flags are required (--dry-run) or forbidden (--force) — i.e., the operator-safety contract — not exact command phrasing.
• TestHealthStateNames only asserts that HEALTH_LABELS.keys() and HEALTH_DESCRIPTIONS.keys() equal the set of six documented state names. It does not inspect the human-readable label values themselves, so renaming "Unbootstrapped" to something else would not break these tests.
• The integration TestAuditIntegration tests use a fake source root with a pinned VERSION file so they're isolated from future version bumps.

No tests are coupled to output formatting or incidental implementation details.