Validation and Testing

Validation And Testing

AgentLogistics uses deterministic local validation plus GitHub Actions checks. The repository is structurally strong, but structural validation is not the same as live model-quality evidence.

Full Local Validation

.\scripts\validate-all.ps1

Expected result:

All AgentLogistics validation checks passed.

Validator Set

The wrapper runs:

• scripts/validate-docs.py
• scripts/validate-taxonomy.py
• scripts/validate-skills.py
• scripts/validate-specializations.py
• scripts/validate-tests.py
• scripts/validate-shared.py
• scripts/validate-skillsets.py

What Local Validation Covers

• required public files, roadmap handoffs, and completion tokens;
• taxonomy naming and duplicate candidate detection;
• skill package frontmatter, sections, references, and agent metadata;
• skillset manifests, route composition, dependencies, exclusions, escalation conditions, fixtures, and role outputs;
• specialization packages, source maps, jurisdiction or mode boundaries, and blocked approval claims;
• scenarios, fixtures, expected routing, evaluation reports, and integration coverage;
• shared formulas, glossaries, schemas, and templates.

GitHub Actions

The Validate workflow runs on pushes to main, pull requests targeting main, and manual dispatch. The Source Link Audit workflow runs weekly and by manual dispatch against specialization source maps.

Both workflows use read-only repository permissions. The source-link audit distinguishes confirmed broken links from transient failures and access restrictions so maintenance review can be targeted.

Current Gap

The repository still needs:

• live model response scoring;
• repeatable source-freshness decisions over time;
• broader numerical fixtures for calculation-heavy skills;
• a final v1 release gate and tag.

See the V1 Release Candidate Audit page and the repository document docs/development/v1-hardening-ci.md for current evidence.