Add WebHarbor task validation script

[Lxr-max]

Summary

This PR adds a small repository-level validator for WebHarbor task files:

• scripts/validate_tasks.py
• scripts/test_validate_tasks.py

It validates the current sites/*/tasks.jsonl schema used in this repo without introducing any new website contribution.

Why this is useful

WebHarbor reviews depend heavily on task quality and consistency. This script gives contributors and reviewers a quick way to catch:

• malformed JSONL
• missing required task fields
• duplicate task IDs within a file or across sites
• localhost/upstream URL mistakes
• task IDs that do not match the current site naming convention
• suspicious task wording, answer-leak heuristics, and bad markers

That makes task review more repeatable and lowers the chance of avoidable PR feedback.

CLI examples

python scripts/validate_tasks.py
python scripts/validate_tasks.py --site amazon
python scripts/validate_tasks.py --tasks sites/amazon/tasks.jsonl
python scripts/validate_tasks.py --strict
python scripts/validate_tasks.py --json

Validation categories

• JSONL parsing and object shape
• Required fields: id, web_name, web, upstream_url, ques
• Duplicate IDs per file and across all sites
• Localhost URL validation for web
• Upstream URL validation for upstream_url
• Port cross-checking against websyn_start.sh
• Suspicious phrases and safety-oriented warnings
• Simple answer-leak and bad-marker heuristics

Checks run

py -m py_compile scripts/validate_tasks.py
py scripts/test_validate_tasks.py
py scripts/validate_tasks.py
py scripts/validate_tasks.py --site amazon
py scripts/validate_tasks.py --tasks sites/amazon/tasks.jsonl
py scripts/validate_tasks.py --strict
py scripts/validate_tasks.py --json

Results:

• py_compile: passed
• self-tests: 6 tests passed
• full-repo validation: 15 sites, 15 task files, 643 tasks, 0 errors, 0 warnings
• --site amazon: passed
• --tasks sites/amazon/tasks.jsonl: passed
• --strict: passed against the current repo state
• --json: returned valid machine-readable JSON summary

Scope notes

• This is not a new website contribution.
• No HF assets are involved.
• .assets-revision was not modified.
• No existing site implementation was changed.
• No site task files were modified.

[jackjin1997]

Claiming this for a reviewer-owned continuation. I will preserve @Lxr-max’s original commit and attribution, and audit the task validator against current main, including schema compatibility, CLI behavior, representative real task files, adversarial positive/negative fixtures, tests, documentation, and integration. This is repository tooling rather than a new mirror, so site UI, task-run, and Hugging Face asset checks are not applicable. I will link a separate Draft Review: PR here before maintainer handoff. I will not merge either PR.

[jackjin1997]

Draft reviewer continuation: #91

The original contributor commit and attribution are preserved. The current candidate has 31/31 tests on Python 3.11/3.12, a strict 24-file / 805-row corpus pass, and 18/18 recorded contract scenarios matching their predefined outcomes. The detailed public report is linked from #91. Independent blind review and reconciliation are still pending, so the Review PR remains Draft. Site UI/HF checks are not applicable to this tooling-only change.