| name | zig-review |
| description | Comprehensive Zig code review across all quality dimensions. Triggers on /zig-review. Runs 7 parallel specialized agents (each on a different model), then a cross-family second-opinion agent on the aggregated result. Always operates on the full branch diff against main. Report-only: never makes changes.
|
| disable-model-invocation | true |
Zig Review
Step 1: Get the diff (for context)
git diff main...HEAD
Fall back to git diff HEAD if that returns nothing.
Step 2: Launch seven parallel review agents
Spawn all seven in a single message so they run concurrently. Each agent fetches the diff itself — do not pass the diff in the prompt. All agents are report-only.
Use these subagent types with the following prompts:
zig-review-tiger-style:
Review the current branch for Tiger Style violations introduced by this PR. Diff against main (git diff main...HEAD). Only flag issues that are new or made worse by this PR — not pre-existing problems in moved code. Report only — no changes.
zig-review-idioms:
Review the current branch for Zig 0.15 idiom and correctness issues introduced by this PR. Diff against main (git diff main...HEAD). Only flag issues that are new or made worse by this PR. Report only — no changes.
zig-review-api:
Review the current branch for public API design issues introduced by this PR. Diff against main (git diff main...HEAD). Pay special attention to new pub declarations that didn't exist on main. Only flag issues introduced or worsened by this PR. Report only — no changes.
zig-review-simplify:
Review the current branch for simplification opportunities in code introduced by this PR. Diff against main (git diff main...HEAD). Only flag complexity that is new to this PR, not pre-existing. Report only — no changes.
zig-review-docs:
Review the current branch for documentation issues introduced by this PR. Diff against main (git diff main...HEAD). Only flag missing/stale docs on newly written or changed functions. Report only — no changes.
zig-review-tests:
Review the current branch for test coverage gaps in code introduced by this PR. Diff against main (git diff main...HEAD). Only flag gaps for new logic or changed behavior — not pre-existing untested code that was merely moved. Report only — no changes.
zig-review-deslop:
Review the current branch for AI-generated comment slop introduced by this PR. Diff against main (git diff main...HEAD). Only flag slop in added lines (+ prefix in the diff). Report only — no changes.
Step 3: Aggregate results
Once all seven agents return, filter before aggregating. Drop any finding that:
- Has no concrete, unambiguous fix ("consider whether...", "you could also...")
- Is already enforced by the compiler or type system
- Is a naming quibble where the existing name is already clear
- Would generate disagreement between two reasonable engineers
- Requires context the reviewer doesn't have
- Would itself generate new findings if acted on (too vague to be actionable)
Produce an intermediate aggregated report in this format:
## Tiger Style
[must-fix findings, then suggestions — or "No issues found."]
## Zig Idioms
[findings — or "No issues found."]
## API Design
[findings — or "No issues found."]
## Simplify
[findings — or "No issues found."]
## Documentation
[findings — or "No issues found."]
## Test Coverage
[findings — or "No issues found."]
## Deslop
[findings — or "No issues found."]
Step 4: Second opinion
Pass the aggregated report to the zig-review-second-opinion agent with this prompt:
Here is an aggregated code review of the current branch, produced by a set of Claude models. Review it independently: challenge weak findings, confirm strong ones, and add anything missed. Fetch the diff yourself for reference.
[aggregated report]
Step 5: Final output
Combine the aggregated primary report with the second opinion into the final output:
## Tiger Style
[findings]
## Zig Idioms
[findings]
## API Design
[findings]
## Simplify
[findings]
## Documentation
[findings]
## Test Coverage
[findings]
## Deslop
[findings]
---
## Second Opinion (GPT-5.5)
### Challenged Findings
[...]
### Confirmed Findings
[...]
### Additional Findings
[...]
### Systemic Observations
[...]
---
**Summary:** X must-fix, Y suggestions across Z categories. Second opinion challenged A findings and added B new ones.
Keep findings concise — one line per issue where possible, with file and line context.