feat: EVALS for data-product-create skill.

[dmitrii-mikh-collibra]

🎯 What does this PR do?

Add Inspect-based eval harness for the data-product-create skill

What

Introduces an evals/ harness that measures whether the collibra/data-product-create skill actually produces the correct asset graph in a real Collibra environment. First eval is the data-product-create happy path; the layout is set up so each future skill gets its own task file.

How it works

The harness uses Inspect AI. A single task drives chip (built as an MCP stdio server) through the full skill flow, then scores the outcome from three independent angles:

• end_state.py — ignores how the agent got there and asks the live environment whether the expected graph exists. Four equally-weighted checks (product exists, ports linked, tables grouped, contract governs a port) so partial results score proportionally.
• trajectory.py — walks the transcript's tool calls to enforce procedural rules:
  • skill_loaded — the skill was discovered/loaded before any write (isolates triggering failures from execution failures).
  • write_order — every create_asset precedes any relation-adding edit, and the contract push comes after its port exists.
• collibra.py — thin async DGC REST 2.0 client shared by the scorers and cleanup.

Other pieces:

• tasks/data_product_create.py — the Inspect task. Injects a system message replicating chip's server-level "discover-and-load-skills" instruction (Inspect doesn't surface it), and supports EVAL_ENABLED_TOOLS to trim the tool surface on low API rate-limit tiers.
• fixtures/data_product_create.yaml — the only per-environment file; holds the source table and expected outcome. The task fails fast with a clear message if unfilled.
• scripts/cleanup.py — deletes eval-created assets between runs (dry-run by default, --apply to execute), since duplicate names break reruns.

✅ Checklist

• My code follows the style guidelines of this project.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation (if needed).
• My commit messages follow the Conventional Commits standard.

[svc-snyk-github-jira]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues
✅	Code Security	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.