Hwaro::Services::Doctor

Every static issue id doctor can emit, derived from CHECK_GROUPS so a new diagnostic can't be validated against a list that forgot it. CHECK_GROUPS is the registry; this is just its index.

missing-config-<section> ids are generated per config section rather than enumerated, so [doctor] ignore validation matches them by prefix.

Sections that are advanced/niche or low-value for most users. These are now treated as opt-in by default (not auto-suggested in normal doctor, and not added by plain --fix unless the user explicitly wants them).

Goal: Reduce config bloat and the "you should add all these things" pressure.

Crinja's FeatureLibrary::UnknownFeatureError message for a tag.

Mirrors MarkdownConfig#initialize defaults — only katex/mathjax render math at runtime; other strings load nothing.

The build's [[content.schema]] check over the regular pages a default build publishes (ContentLister, plus [[content.generate]] pages), with section cascades resolved by the build's own build_cascade_map / merged_cascade_for. {file, result} per checked page, in path order. Shared by doctor and tool validate.

The scaffolded placeholder titles. Models::Config falls back to "Hwaro Site" when title is absent, and every site hwaro init creates ships its scaffold's placeholder — so checking only the internal fallback meant the advisory never fired on the sites that actually have a placeholder title, and it shipped into <title>, OG tags and feeds unnoticed.

Sourced from the scaffold registry rather than a hand-kept literal list: hardcoding "My Hwaro Site" silently exempted every --scaffold blog|docs|book site, which ships "My Blog" / "My Docs" / "My Book". A new scaffold now joins the check automatically.

True for any id [doctor] ignore can legitimately name.

Apply real fixes (Phase 1: value corrections like base_url trailing slash) and optionally approve/add recommended config sections (Phase 2).

• apply_value_fixes: Phase 1 value corrections. --fix and --full enable this; a bare --approve must NOT edit real values — the documented model is "--fix normalizes, --approve adds sections".
• approve_sections: When true, doctor will add the recommended/optional config sections as commented documentation (--approve / --full).

This separation makes --fix focused on corrections, while --approve / --full controls bringing in the larger set of recommendations.

Returns the list of config section keys missing from the user's config.toml. On I/O or TOML parse failure, returns an empty array — those paths are already reported by check_config as classified issues, so silent-empty here lets the main run loop carry on without double-reporting. Callers that need a clearer signal (e.g. fix_config) should probe the file directly via readable_config_toml / parse_config_toml.

The blocking ids the last run observed, recorded BEFORE the [doctor] ignore filter. The CLI's skip rule keys off this rather than off the returned issues: content-dir-missing is a warning, so ignore = ["content-dir-missing"] used to hide it AND take the whole content/ group's skip state with it — every check in the group went back to rendering ✓ for a site with no content directory at all, which is the false all-clear this mechanism exists to prevent. Silencing a report is not the same as the scan having run.