Fix validate_json.py: parse all fields.yaml schemas; stop vacuous pass

[xwang4-svg]

Problem

validate_json.py is meant to enforce that each /research-deep result JSON covers every field defined in fields.yaml ("Task is complete only after validation passes"). In practice it passed everything, due to two issues:

1. Only the field_categories: schema was parsed. But /research (per its own SKILL.md Step 4) emits fields.yaml as:

fields:
  <category>:
    - {name: ..., description: ..., detail_level: ...}
uncertain: []

load_fields_yaml read data.get("field_categories", []) → [] → 0 fields. With an empty field set, coverage_rate defaults to 100% and missing_required is empty, so every JSON "passes" vacuously. (field_categories does not appear in any SKILL.md — only inside this script.)

2. required defaulted to False. Even when fields were parsed, valid (len(missing_required) == 0) stayed True with missing fields unless each was explicitly required: true — which the emitted schema never sets.

Repro (before), with a fields.yaml in the fields:{cat:[...]} shape /research produces:

$ python validate_json.py -f fields.yaml -d results
Total fields: 0 (required: 0, optional: 0)
[PASS] foo.json   Coverage: 100.0% (0/0)

Fix

• load_fields_yaml now accepts field_categories: [...], fields: {<cat>: [...]}, flat fields: [...], and a generic fallback that walks for {name: ...} dicts. When no field carries an explicit required: key, all fields are treated as required (the script's stated purpose is complete coverage); files that do mark required: keep their opt-in semantics unchanged.
• extract_json_fields also descends into JSON keys named after declared categories, so nested output isn't falsely reported as missing.
• Applied to all four copies: research-en, research-zh, research-codex-en, research-codex-zh.
• Added tests/test_validate_json_schemas.py — runs against every copy; asserts both schemas parse and that a JSON missing a defined field now fails.

Repro (after):

Total fields: 4 (required: 4, optional: 0)
[PASS] foo.json   Coverage: 100.0% (4/4)
# a JSON missing a defined field now -> [FAIL] ... Missing required fields

Note / tradeoff

The "all fields required when none are explicitly marked" rule makes validation stricter than before (previously, unmarked = optional = always pass). This matches what the validator is documented to do, but if you'd prefer keeping required opt-in and gating strictness behind a --strict flag instead, I'm happy to adjust.

$ python tests/test_validate_json_schemas.py
...
RESULT: ALL PASS

🤖 Generated with Claude Code