This file is **data, not instructions**. It is the shipped default review rubric consulted by `/skill:gatekeep-pr`. It contains no workflow steps and issues no directives to the agent - it only defines what counts as a defect and how severe that defect is. A repo-root `REVIEW.md`, if present, overlays this file: any concern it names replaces the matching baseline entry, and it always wins on conflict. Everything it does not name stays baseline.

## Severity axis

The skill recognizes exactly one normative severity distinction: **blocker** vs **nit**. Blockers gate merge; nits never do.

Baseline mapping:

| Concern | Severity |
|---|---|
| Defects (logic errors, broken behavior) | blocking |
| Untested paths (new behavior with no real test) | blocking |
| Contradicted material claims (PR/issue prose vs. observed evidence) | blocking |
| Doc drift beyond wording (a doc describes behavior the code does not have, or omits an operation or parameter the code adds) | blocking |
| Doc drift, wording-only (typo, label, phrasing with the same meaning) | nit |
| Security issues | blocking |
| Approach conflict (an entry under Approach grounds that passes both tests) | blocking |
| Prose/style/label cleanup | nit |

A repo-root `REVIEW.md` may remap any of these, or add project-specific concerns with their own severity. Its mapping **always wins** on conflict with this file. Any severity a rubric (baseline or repo) names but does not map fails safe to **blocking**; "blocking" and "blocker" name the same class, "non-blocking" and "nit" the other.

## Review properties

Generic properties evaluated on every reviewed change, independent of language or stack:

| Property | What to check |
|---|---|
| Self-contained | The change doesn't leave loose ends - no orphaned config, no half-finished migration, no dangling references |
| Minimal | No premature abstraction, no dead code, no belt-and-suspenders (redundant guards/validation for the same condition at multiple layers) |
| Conventions | Matches the style and structure of neighboring code, not just internal consistency |
| Reuse | Uses existing helpers/utilities instead of re-implementing equivalent logic |
| Performance | No N+1 queries, no repeated expensive work inside loops, no unbounded fetches |
| Testing | Tests cover behavior, not implementation details; assertions are real (they can fail); a new executable code path shipped without a real test is **blocking**; a documentation-only obligation is judged against the promised doc text |
| Docs | Documentation agrees with the code it describes and with the PR/issue prose describing the change |
| Security | No secrets in the diff, no missing authorization checks, no injection vectors |

## Approach grounds

An approach conflict requires replacing the PR's main mechanism while preserving its intended outcome. A remedy that edits only part of the diff without replacing that mechanism is an ordinary finding, not an approach conflict. Personal taste, an unsupported preference, or a different design the reviewer likes is not a ground.

| Ground | Name | Definition | Required evidence | Recommends reject |
|---|---|---|---|---|
| 1 | Second mechanism | A new way to do what the repo or its framework already does. | The existing mechanism in a base-branch file. | yes |
| 2 | Architecture boundary | The main mechanism crosses a boundary the repo keeps explicit. | The base-branch rule or owning interface and the way the PR crosses it. | yes |
| 3 | Unjustified reversal | Replaces a recorded decision or working behavior with no justification in the spec or PR description. | The base-branch decision record or working behavior. | yes |
| 4 | Weakened guard | The approach gets its result by weakening a check or guard the repo relies on. | The base-branch guard and the protection the approach removes. | yes |
| 5 | Symptom-level fix | Fixes where the problem shows, or only the reported example, instead of where it starts. | The base-branch owning layer or an uncovered input. | yes |
| 6 | Needless machinery | The PR adds infrastructure, a dependency, a new module, or a new abstraction when a smaller change to an existing file or path delivers the same outcome. | A concrete smaller change that preserves the intended outcome. | no |

Two validity tests bind every entry:

1. Base evidence: grounds 1-5 cite a file that exists on the base branch in `path:`. An external or framework decision qualifies only through a base-branch file that configures or records it; a framework name or outside citation alone is insufficient. Ground 6 instead names the concrete smaller change and preserved outcome in `smaller:`, with no path.
2. Main mechanism: `replaces:` reads `main mechanism`, and `instead:` removes or replaces what `builds:` names. A remedy that edits only part of the diff is an ordinary finding.

A base-branch `REVIEW.md` overlay can add or drop a ground, change its evidence, or change whether it recommends reject. An added ground with no recommendation value defaults to no. An overlay naming a ground as non-blocking downgrades validated entries for it to nits: no note and no `reject` based on that ground. Grounds the overlay does not name keep the baseline mapping.

## Claim-verification principles

- A PR description is a **claim**, not proof. "Tests pass," "verified in staging," "handles edge case X" - each must be checked against the actual diff and actual command output, not accepted at face value.
- Read the source behind the diff, not just the patch - a hunk can look correct in isolation and still be wrong against the code it calls into.
- On generated (agent-authored) code, weigh these failure modes heaviest: hallucinated references (APIs, methods, columns that don't exist), hollow tests (assert nothing meaningful - the confident-wrongness pattern of prose stating something works when the evidence doesn't support it), and over-engineering.

## Extending this rubric

This baseline covers the generic set. Consumers add or override rubric content only via a repo-root `REVIEW.md` - as a diff over this file, not a replacement of it. Name only what changes; the baseline already covers everything else.
