Large diffs hide bugs
Coding agents generate hundreds of lines. Skimming is how production issues land.
Loading home
agent PR review · markdown rubrics
The problem
Cursor, Claude Code, and Codex generate hundreds of lines. Without a review structure, PRs get merged with secrets in logs, fake APIs, and hollow tests. Spec Gate is five rubrics, two checklists, and PR templates. Paste scores into GitHub review comments.
Failure patterns
Coding agents generate hundreds of lines. Skimming is how production issues land.
Mocked endpoints and hollow test coverage look green until production traffic arrives.
Without a rubric, reviews default to 'looks fine' or block the entire PR over style.
Five rubrics
Each rubric has a 5min skim path and a 20min deep path. Use the smallest review that closes your current risk.
Secrets in logs, hardcoded credentials, missing validation, and agent-specific exposure patterns.
Hollow mocks, untested edges, missing error paths, and test coverage that claims more than it proves.
Breaking changes without migration path, schema drift, and endpoint contracts that skip versioning.
N+1 queries, unbounded loops, model token waste, and API cost patterns agents introduce.
Commented-out debugging, orphaned imports, invented metrics, and hallucinated dependencies.
Right fit
Use Spec Gate when you're the reviewer and the agent opened the PR. Skip it if you want automated LGTM without reading the diff.
Founding offer
Checkout charges $49 once, in USD. After payment you get the verified v1.0.0 ZIP with 5 rubrics, 2 checklists, and PR templates, plus the recovery email link.
$49USD, one time. Provided by SplitLabs OÜ. The checkout is sold through Link, Stripe's Managed Payments merchant of record. Immediate private ZIP delivery. 14-day consumer withdrawal where applicable. Terms.
Immediately after Stripe confirms the purchase. nyk.dev verifies the live order and serves the checksum-matched v1.0.0 ZIP containing 5 rubrics, 2 checklists, and PR templates. A recovery link is also emailed.
Manual review. Spec Gate is a markdown pack for humans. No bot, no account, no automated LGTM. The rubrics guide you through scoring the PR before merge.
Claude Code, Codex, Copilot, and similar tools. The patterns apply to any agent generating code diffs, not just one host.
The $49 license is for one person. Team redistribution is not included. Team demand will inform a separate license.
Spec Gate scores agent PRs. If you want the full control layer that decides what the agent may do in the first place, Production Agent OS Kit is the one to read next — €149.