Name: loona-test-project
Author: Zhaozed

Imported from Zhaozed/Loona-PMdevelopskill (skills/loona-test-project/SKILL.md). Install upstream with npx skills add Zhaozed/Loona-PMdevelopskill --skill loona-test-project. Copyright stays with the author.

Loona Intent Testing

Use the bundled Rust tester for repeatable regression tests after every model training.

API

Default endpoint is configured in loona-intent-tester/src/config.rs:

http://117.50.221.35:35000/classify

Request: { "text": "..." }.

Build

cd /Users/joey/.codex/skills/loona-test-project/loona-intent-tester
cargo build --release

Binary:

/Users/joey/.codex/skills/loona-test-project/loona-intent-tester/target/release/loona-intent-tester

Standard full-language test

Use the final multilingual workbook directory. Start with --dry-run, then run the API test:

/Users/joey/.codex/skills/loona-test-project/loona-intent-tester/target/release/loona-intent-tester \
  --input-dir "/path/to/corpora" \
  --all-languages \
  --output-dir "/path/to/corpora/test_reports" \
  --concurrency 20 \
  --threshold 0.85 \
  --max-retries 3

For only formal test sheets, add repeated --include-sheet options. For one language use --language thai (aliases such as th and 泰语 are supported).

Modes

--dry-run: detect language/intent columns and count extracted tasks without API calls.
--include-file TEXT: test only filenames containing TEXT; repeatable.
--include-sheet NAME: test exact sheet names; repeatable.
--sheet-contains TEXT: test sheets whose names contain TEXT; repeatable.
--retry-api-failures-from REPORT.xlsx: retry only API failure rows.
--resume CHECKPOINT.json: resume an interrupted run.
--retest-included-sheets: retest selected sheets and merge into the latest language reports.
--use-existing-reports: reuse existing latest reports when possible.
--rebuild-existing-reports: rebuild report layout without API calls.

Report interpretation

Each row is classified as:

完全正确: predicted label matches true label and score meets threshold.
正确但置信度不足.
错误但置信度高: highest-priority model/label-boundary issue.
错误且置信度低.
API请求失败.

Reports are written as intent_<language>_<timestamp>.xlsx and intent_overall_<timestamp>.xlsx, containing summary, detailed results, per-sheet summaries, and error-category sheets.

Regression-test policy

For every newly trained model, run all supported languages on the same frozen test set. Track overall accuracy, per-language accuracy, per-intent recall, high-confidence error count, API failure rate, and latency. Compare against the previous report before approving the model.

Do not use the legacy scripts/test_loona_tables.py unless its endpoint and payload are updated; it contains an older API configuration.

Install loona-test-project and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "zhaozed-loona-pmdevelopskill-loona-test-project", "kind": "skill", "name": "loona-test-project", "description": "Test Loona multilingual BERT intent classification from Excel corpora against the deployed /classify API. Use when a trained Loona model needs full-language regression testing, per-language accuracy reports, confidence/error analysis, API-failure retry, checkpoint resume, or retesting selected sheets.", "publisher": "Zhaozed", "version": "1.0.0", "capabilities": { "domains": [ "coding", "hr" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Test Loona multilingual BERT intent classification from Excel corpora against the deployed /classify API. Use when a trained Loona model needs full-language regression testing, per-language accuracy reports, confidence/error analysis, API-failure retry, checkpoint resume, or retesting selected sheets." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/Zhaozed/Loona-PMdevelopskill", "path": "skills/loona-test-project/SKILL.md", "ref": "6b090e4418bde3af7756a01a2043c944f967e27b", "url": "https://github.com/Zhaozed/Loona-PMdevelopskill/blob/6b090e4418bde3af7756a01a2043c944f967e27b/skills/loona-test-project/SKILL.md", "key": "Zhaozed/Loona-PMdevelopskill/skills/loona-test-project/SKILL.md" } }, "instructions": "# Loona Intent Testing\n\nUse the bundled Rust tester for repeatable regression tests after every model training.\n\n## API\n\nDefault endpoint is configured in `loona-intent-tester/src/config.rs`:\n\n```text\nhttp://117.50.221.35:35000/classify\n```\n\nRequest: `{ \"text\": \"...\" }`.\n\n## Build\n\n```bash\ncd /Users/joey/.codex/skills/loona-test-project/loona-intent-tester\ncargo build --release\n```\n\nBinary:\n\n```text\n/Users/joey/.codex/skills/loona-test-project/loona-intent-tester/target/release/loona-intent-tester\n```\n\n## Standard full-language test\n\nUse the final multilingual workbook directory. Start with `-", "cost": { "context_tokens": 652 } }

loona-test-project

About

Loona Intent Testing

API

Build

Standard full-language test

Modes

Report interpretation

Regression-test policy

Install loona-test-project and it becomes one more candidate the router can pick - when it fits.

Use it

Manifest

Reviews