| name | improve-code-quality |
| description | Analyze and improve code quality for any path in the monorepo. Use whenever the user asks to improve, audit, review, clean up, refactor, lint, or check code quality for a directory or file. Also trigger for requests about security review, performance optimization, robustness checks, dependency audits, or readability improvements. Activate for any mention of code quality, code health, code smells, tech debt, or cleanup tasks targeting a specific path. Even if the user just says "check this code" or "review crates/foo", this skill applies. |
Improve Code Quality
Perform a structured code quality audit on a target path in the monorepo. The workflow is: run
automated tools, do deeper manual analysis, present a report, then apply fixes only after the user
approves.
Usage
/improve-code-quality <path>
The path can be a directory (e.g., crates/task-runner), a single file (e.g.,
crates/task-runner/src/lib.rs), or the name of a package (e.g., proto_core).
Step 1: Validate and detect language
First, verify the target path exists. If it doesn't, tell the user and stop.
Then detect what languages are present:
- Rust: path is under
crates/, plugins/, or contains .rs files or a
Cargo.toml
Tell the user what you detected before continuing.
If the target is a single file, skip the automated tool phase and go straight to manual analysis.
Step 2: Manual analysis
Read the source files in the target path and look for issues that automated tools miss. Organize
findings by category:
Security
unsafe blocks without justification comments
.unwrap() on user-facing code paths (not tests — unwrap is fine in tests)
- Path traversal or injection risks
- Hardcoded secrets or credentials
- Use of
std::collections::HashMap or std::collections::HashSet (refer to conventions below)
Performance
- Unnecessary
.clone() where borrowing would work
- Unnecessary
.collect() in iterator chains
- O(n^2) or worse algorithms with better alternatives
- Blocking calls inside async functions
- Large types on the stack that should be
Boxed
- Avoid cloning in loops; use
.iter() instead of .into_iter() for Copy types
- Prefer iterators over manual loops (when applicable); avoid intermediate
.collect() calls
- Detect memory leaks or dangling pointers
Readability & structure
- Functions longer than ~80 lines that should be split
- Deeply nested or complex match/if-else chains
- Dead code or unused imports
- Inconsistent naming
- Magic numbers that should be named constants
Robustness
- Missing error handling (bare
.unwrap() in non-test production code)
- Incomplete pattern matches
- Missing input validation at boundaries
- Race conditions in concurrent code
Borrowing & ownership
- Prefer
&T over .clone() unless ownership transfer is required
- Small
Copy types (≤24 bytes) can be passed by value
- Use
Cow<'_, T> when ownership is ambiguous
Dependencies
- Unused dependencies in
Cargo.toml
- Overly broad feature flags on dependencies
Step 3: Run automated tools
Run the appropriate tools from the repository root (without asking for permission if possible).
Capture all output, including errors.
Find the crate name by reading the Cargo.toml in the target directory (the [package] name
field). If the target is a subdirectory within a crate, walk up to find the crate root.
cargo check -p <crate_name> 2>&1
cargo clippy -p <crate_name> --all-targets -- -D warnings 2>&1
cargo fmt -p <crate_name> --check 2>&1
cargo nextest run -p <crate_name> --no-fail-fast -j 4 2>&1
For crates under plugins/, run from the plugins/ directory instead.
If a tool fails
If clippy can't run because of build errors, capture those errors — they become the highest priority
findings. Don't abort the rest of the analysis.
Large scope warning
If the target is the repo root or a very broad path, warn the user it will take a while and suggest
narrowing scope.
Step 4: Present the report
Show all findings in a structured report:
# Code quality report: <target_path>
**Language(s):** Rust
**Files analyzed:** <count>
---
## Automated tool results
### Checking
<summarize warnings/errors>
### Linting
<summarize warnings/errors with file:line references>
### Formatting
<list files with issues, or "All files properly formatted">
### Testing
<list failing tests>
---
## Manual analysis
### Critical — Security
| # | File:Line | Finding |
|---|-----------|---------|
| 1 | path/file.rs:42 | Description |
### High — Performance
| # | File:Line | Finding |
|---|-----------|---------|
### Medium — Readability & Structure
| # | File:Line | Finding |
|---|-----------|---------|
### Low — Robustness
| # | File:Line | Finding |
|---|-----------|---------|
### Info — Dependencies
| # | File:Line | Finding |
|---|-----------|---------|
---
## Summary
- Critical: <n> | High: <n> | Medium: <n> | Low: <n> | Info: <n>
- **Total: <n>**
## Recommended fix order
1. <highest priority>
2. ...
If a category has no findings, show the header with "No issues found."
Severity guide:
- Critical: Security vulnerabilities, correctness bugs, data loss risks
- High: Performance problems, error handling gaps with production impact
- Medium: Readability issues, structural problems, maintainability concerns
- Low: Minor style issues, documentation gaps
- Info: Dependency notes, suggestions, non-actionable observations
Step 5: Apply fixes (after approval)
After showing the report, ask: "Which findings should I fix? (all / critical+high / specific
numbers / skip)"
Wait for the user's response. Do not apply anything without explicit approval.
Fixing process
-
Run auto-fix tools first:
- Linting:
cargo clippy -p <crate> --all-targets --fix --allow-dirty --allow-staged
- Formatting:
cargo fmt -p <crate> -- --emit=files
-
Apply manual fixes one at a time, stating what changed and why for each.
-
Re-run automated tools to verify no regressions were introduced.
-
Do not commit. Tell the user the changes are ready for their review.
Step 6: Validate best practices
After applying fixes, check that the target path adheres to best practices by running the
/rust-skills skill. If this skill does not exist, skip this step.
If you notice any new issues or deviations from conventions, add them to the report and ask the user
if they want to address those as well.
proto specific conventions
Always enforce these project rules during analysis:
- No
std::collections::HashMap/HashSet — use rustc_hash::FxHashMap/FxHashSet
- Exception when using a third-party crate that requires
std collections in its public API
(e.g., indexmap), but even then internal code should prefer FxHashMap/FxHashSet
- Rust edition/toolchain idioms
- Infer toolchain from
rust-toolchain.toml
- Unix newlines for Rust files
cargo-nextest for running tests (cargo nextest, not cargo test)