Refactor ECNU API skill for safe execution and reproducible validation

[JJasonSun]

Summary

The repository provides a task-oriented ECNU API Agent Skill with implementation, review, troubleshooting, live-verification, privacy, cost, and provenance workflows.

Behavior

• classify requests before execution and load only the references needed for the task
• validate protocol roots, models, request fields, timeouts, response shapes, privacy boundaries, and retry safety
• send LangChain embedding requests without an unsupported dimensions field and verify the returned 1024-value vector
• separate official API contracts from dated live observations in references/known_deviations.md
• provide fallback guidance for inconclusive /models responses and ecnu-max[1m] metadata or authentication errors
• provide focused implementation, review, troubleshooting, retry, model-discovery, embedding, Anthropic, security, and reporting procedures
• provide a sanitized standard-library smoke test with model-list checks by default and explicit opt-in low-cost POST and Anthropic profiles
• redact secrets and successful model output from structural smoke-test reports
• validate required files, frontmatter, relative links, key-shaped literals, machine-specific paths, and embedding request fields offline
• run repository validation and unit tests in GitHub Actions

Validation

• python scripts/validate_skill.py — passed
• python -m unittest discover -s tests -v — 5 tests passed
• Python syntax compilation — passed
• mock HTTP integration for scripts/smoke_test.py — passed