Add DataJunction deep code-review skill

[robinld]

Summary

Add a maintainer-focused review skill for DataJunction changes. It traces semantic-layer correctness, query grain, materialization, metadata, authorization, API/client compatibility, and lifecycle risks beyond the changed lines.

The automated-review reference defines a structured handoff for an advisory review bot, including exact-head verification, thread reconciliation, and a single maintained summary. It does not authorize the skill user to publish comments.

Review feedback

Greptile's initial summary reviewed 5ce0e561, not the current head. Its three findings were addressed in d46692f5: structured reviews now require a trusted discussion snapshot; the strict handoff schema requires the publisher's verdict, actions, and test gaps; and thread decision/index combinations are constrained in that schema and validated before writes. Replies with details are in the original review threads. The later 17ed7f63 commit adds a focused rule to trace new invariants through affected paths.

Verification

The skill bundle matches the reviewer and worker policy copies in the internal pilot; the skill validator and schema checks passed. No application runtime behavior changes.

[netlify]

✅ Deploy Preview for thriving-cassata-78ae72 canceled.

Name	Link
🔨 Latest commit	17ed7f6
🔍 Latest deploy log	https://app.netlify.com/projects/thriving-cassata-78ae72/deploys/6abcd05f58d20000070cb0e0

[greptile-apps]

Confidence Score: 4/5

[Low risk] Adds documentation and schemas for a code review skill.

The PR should not merge until structured reviews without a trusted discussion snapshot have a valid handoff path.

Findings

1.  Digest Required Without Snapshot ▶
2.  Publishing Fields Not Required ▶
3.  Conflicting Thread Actions Accepted ▶

Summary

Adds a maintainer-oriented DataJunction review skill and an automated-review handoff with generation and publishing schemas. The review guidance is detailed, but the structured handoff needs a defined snapshot-less path and tighter agreement between its two schemas.

• Align digest requirements with the inputs available to an orchestrated review.
• Make required publishing fields and thread-action constraints enforceable.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Orchestrator inputs] --> B[Review skill]
  B --> C[Model-schema generation]
  C --> D[Handoff-schema validation]
  D --> E[Publisher]
  A -. Trusted discussion snapshot, if supplied .-> C

Loading

_{Reviews (1) · Last reviewed commit: "Add DataJunction adversarial review skil..."}