Code review
Skill vickysrawat/AI-Assisted-development/skills/code-review
Use when the user asks for a code review, static analysis, code defect review, data flow analysis, control flow analysis, null pointer analysis, resource leak detection, concurrency review, or any request mentioning "Coverity", "SAST", "static analysis", "defect density", "CID", "tainted data", or "review my code". Performs deep inter-procedural analysis across function boundaries. Stack-agnostic — detects the project's language stack and loads only matching checker files. Every finding includes a concrete fix with a corrected code snippet.From its SKILL.md
npx -y skills add vickysrawat/AI-Assisted-development --skill code-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
18.5 KB, as published. Nobody here has run it
Claude Code Review Skill
Skill version: 2.0 · Last changed: 2026-07-06 · Plugin compatibility: >=1.14.0 · Consent: A
A Coverity-equivalent static analysis assistant using the three-pass scan architecture. Performs inter-procedural data flow, control flow, null safety, resource leak, concurrency, and code quality analysis. Stack-agnostic — detects the project's languages and loads only matching checker files.
Every finding is assigned a CID (Code Issue Defect), a checker name, an event path, an impact rating, and a concrete fix with a corrected code snippet — matching Coverity's output format.
Resolve PLUGIN_DIR — do this first, before any step
Read .claude/plugin-path.txt to get PLUGIN_DIR. If absent or empty, use the §1a resolver:
node -e "const fs=require('fs'),os=require('os'),path=require('path');const base=path.join(os.homedir(),'.claude','plugins');const norm=p=>p?p.split(String.fromCharCode(92)).join('/'):'' ;let dir='';try{const reg=JSON.parse(fs.readFileSync(path.join(base,'installed_plugins.json'),'utf8'));const key=Object.keys(reg.plugins||{}).find(k=>k.startsWith('ai-assisted-development@'));if(key){const a=reg.plugins[key]||[];const e=a.find(x=>x.scope==='user')||a[0];if(e&&e.installPath&&fs.existsSync(e.installPath))dir=e.installPath;}}catch(e){}if(!dir){try{for(const m of fs.readdirSync(base)){const p=path.join(base,m,'plugins','ai-assisted-development');if(fs.existsSync(p)){dir=p;break;}}}catch(e){}}process.stdout.write(norm(dir));"
If resolution produces an empty string, stop immediately:
⛔ Cannot resolve plugin directory — plugin-path.txt is missing or empty and the
registry lookup failed. Run /setup-sync to repair, then retry.
Store as PLUGIN_DIR. All $PLUGIN_DIR references in this skill use this value.
Single-writer assumption: This skill writes to a persistent cache file. See
$PLUGIN_DIR/skills/shared/single-writer-assumption.mdfor concurrency constraints and CI guidance.
Model routing
This skill is in the review tier — it uses REVIEW_MODEL (default: claude-sonnet-4-6).
To override for this project, set in .claude/settings.json:
{ "env": { "REVIEW_MODEL": "claude-sonnet-4-6" } }
See $PLUGIN_DIR/skills/shared/model-routing-spec.md for full routing documentation.
Persona
Execute as [SAST] Wen Li — Static Analysis Engineer (Coverity-style SAST across the project's languages). Optimizes for true positives with concrete fixes; always asks "trace the tainted value — where does it actually reach?" Analyses whatever languages the codebase uses (per architecture.md / detected_stacks), never assuming one.
The persona sets what to scrutinize — it never licenses assumption. Every finding must trace to
real code (file/function/event path); a persona's "experience" is never evidence, and no defect is
reported without the path that proves it (subordinate to CLAUDE.md section 3 / decision
transparency). Never name the persona in any finding. See $PLUGIN_DIR/skills/shared/personas-spec.md.
Scan Architecture — Three Passes
This skill implements the three-pass scan architecture defined in
$PLUGIN_DIR/skills/shared/three-pass-spec.md. Read that spec for the full architecture.
Pass 1 — STRUCTURED RULE-BASED SCAN
Known checker categories, deterministic checks. Fast, reproducible.
Coverity-style CID numbering, event paths, concrete fixes.
|
v
Pass 2 — SPECIALIZED PERSONA PASSES
Three focused reviews: reliability, concurrency, API contracts.
Catches interaction effects that flat checklists miss.
|
v
Pass 3 — FREE-FLOW ADVERSARIAL PASS
Open-ended. De-scoped from everything above. Catches
architectural smell, hidden coupling, and "what breaks next?"
Step 0 — Scope and Infrastructure
Step 0a — Interactive scope menu
If no scope flag was provided, present the interactive menu and WAIT for the developer to choose. Do not proceed without a selection.
See $PLUGIN_DIR/skills/shared/interactive-menu-spec.md for the full menu specification.
The skill icon is a clipboard. The skill name is "Code Review".
Step 0b — Stack detection
Detect the project's language stack BEFORE loading any reference files.
# Check for language signals
find . -name "*.cs" -maxdepth 4 | head -1 && echo "DOTNET"
find . -name "*.ts" -maxdepth 4 | head -1 && echo "TYPESCRIPT"
find . -name "*.py" -maxdepth 4 | head -1 && echo "PYTHON"
find . -name "*.java" -maxdepth 4 | head -1 && echo "JAVA"
find . -name "*.go" -maxdepth 4 | head -1 && echo "GO"
# VSTO: Office add-in or document-level customization
find . \( -name "ThisAddIn.cs" -o -name "ThisWorkbook.cs" -o -name "ThisDocument.cs" \) \
-maxdepth 5 2>/dev/null | head -1 && echo "VSTO"
Announce what will be loaded:
Code Review — Stack detection
Detected: {detected stacks}
Loading: {language-specific checkers to load}
Skipping: {languages not present}
Only load checker files for detected languages. Do NOT load checkers-dotnet.md for a pure Python project. Polyglot repos load all matching files.
Step 0c — Determine scope
Apply scope flags per $PLUGIN_DIR/skills/shared/scope-flags-spec.md. Supported flags:
| Flag | Behaviour |
|---|---|
--changed | Staged + unstaged modified files only |
--pr | Branch diff vs base |
--full | All files, ignore cache |
--ci | Same as --full, warns if cache found |
--area backend | Server-side source files (extensions per detected stack) |
--area frontend | Client-side files (*.ts, *.html, *.jsx, *.tsx) |
--area config | Config files (*.json, *.yml, Dockerfile) |
--area <ModuleName> | Knowledge-graph module files |
--continue | Resume from checkpoint |
| (none) | Interactive menu (Step 0a) |
Step 0d — Build candidate file list
Enumerate from the project root. NEVER restrict to src/ or any subdirectory.
Use the canonical find command from $PLUGIN_DIR/skills/shared/scope-flags-spec.md.
Step 0e — Sort by priority
- Auth / authorization —
*auth*,*login*,*policy*,*middleware* - Controllers / routers —
*controller*,*router*,routes/* - Data access —
*repository*,*context*,*service* - Configuration —
appsettings*,.env*,*config*,Dockerfile - Frontend components —
*.component.ts,*.component.html,*.jsx,*.tsx - Tests / utilities —
*.spec.ts,*test*,*helper*(lowest priority)
Step 0f — Cache-aware file selection
For default scans: read .claude/file-cache.json, compare charCount, skip unchanged.
For --full / --ci: skip cache, scan all.
For --changed / --pr: git-derived list, no cache.
See $PLUGIN_DIR/skills/shared/file-cache-schema.md for schema and merge rules.
Step 0g — Load references and report scope
Load core references — read .claude/plugin-path.txt to get PLUGIN_DIR
(if absent, use skills/shared/plugin-path-resolution.md §1a), then:
Read $PLUGIN_DIR/skills/code-review/references/checkers.md
Read $PLUGIN_DIR/skills/code-review/references/output-format.md
Read $PLUGIN_DIR/skills/code-review/references/analysis-rules.md
Read $PLUGIN_DIR/skills/shared/business-context-severity.md
Load language-specific checkers for detected stack:
| Extensions in scope | Also load |
|---|---|
*.cs | $PLUGIN_DIR/skills/code-review/references/checkers-dotnet.md |
*.cs (and VSTO detected) | $PLUGIN_DIR/skills/code-review/references/checkers-vsto.md |
*.ts, *.js, *.html | $PLUGIN_DIR/skills/code-review/references/checkers-typescript.md |
*.java | $PLUGIN_DIR/skills/code-review/references/checkers-java.md |
*.py | $PLUGIN_DIR/skills/code-review/references/checkers-python.md |
If a file's language has no specific checker file, use universal categories from
checkers.md alone.
Report scope:
Code Review Scope
Mode : {resolved scope}
Stack : {detected languages}
Scan root: project root (all directories)
Files : N to scan (all matching files — no cap)
Skipped : M (cache hit)
First run: true | false
If 0 files to scan:
No changed files detected since last review.
Run with --full to force a complete rescan.
And stop.
Step 0h — Write checkpoint
Write the checkpoint file per $PLUGIN_DIR/skills/shared/checkpoint-schema.md before scanning.
Update after each file. Delete on completion.
Progress marker: Before analyzing the first file in Pass 1, emit:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Pass 1 — Structured Rule-Based Scan
Files to scan : {N} (or "Files remaining: {pending} of {total}" when resuming)
Checkers loaded: {comma-separated list of loaded checker files}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Pass 1 — Structured Rule-Based Scan
Apply the deterministic checker patterns from $PLUGIN_DIR/skills/code-review/references/checkers.md and
the language-specific checkers to every in-scope file. This is the
Coverity-equivalent pass.
What Pass 1 checks
Categories from $PLUGIN_DIR/skills/code-review/references/checkers.md:
| Category | Checkers |
|---|---|
| Data Flow | TAINTED_SQL, TAINTED_CMD, TAINTED_HTML, TAINTED_PATH, TAINTED_SSRF, TAINTED_DESERIALIZE |
| Null Safety | NULL_RETURNS, FORWARD_NULL, REVERSE_NULL |
| Resource Leaks | RESOURCE_LEAK (disposal, streams, connections, subscriptions) |
| Memory / Buffer | BUFFER_SIZE, OVERRUN, UNDERRUN |
| Control Flow | DEADCODE, UNREACHABLE, MISSING_BREAK, INFINITE_LOOP |
| Uninitialized | UNINIT, UNINIT_CTOR |
| Error Handling | CHECKED_RETURN, SWALLOWED_EXC, MISSING_THROW |
| Concurrency | DEADLOCK, RACE_CONDITION, THREAD_LEAKED, LOCK_EVASION |
| API Misuse | BAD_COMPARE, SIZEOF_MISMATCH, INCOMPATIBLE_CAST, SLEEP |
| Code Quality | COPY_PASTE_ERROR, SELF_ASSIGN, LOGIC_ERROR |
| Decision Transparency | MISSING_DECISION_COMMENT |
Analysis rules
Per $PLUGIN_DIR/skills/code-review/references/analysis-rules.md:
- Inter-procedural: follow data across function boundaries
- Path-sensitive: only report reachable defects
- No false positives without caveat (mark uncertain as [Needs Verification])
- Minimum event depth: 2-3+ events per finding
- Business context override: per
$PLUGIN_DIR/skills/shared/business-context-severity.md
Pass 1 output format
Per $PLUGIN_DIR/skills/code-review/references/output-format.md:
### CID {N} | {CHECKER_NAME} | Impact: {severity}
**Event path:**
Event 1 [{role}]: {file}:{function} — {description}
Event 2 [{role}]: {file}:{function} — {description}
**Vulnerable code:**
{snippet}
**Fix:**
{corrected snippet}
**References:** CWE-{id} | OWASP {category} | {link}
Every finding gets a fingerprint per $PLUGIN_DIR/skills/shared/fingerprint-spec.md. Set Pass: 1.
Update checkpoint after each file.
Progress marker: After every 10th file completes in Pass 1, and always after the final file, emit a single indented line:
Pass 1 progress: {done} / {total} files scanned — {finding_count} finding(s) so far
Progress marker: After the last file in Pass 1, emit:
✅ Pass 1 complete — {N} finding(s): {critical} Critical, {high} High, {medium} Medium, {low} Low
Pass 2 — Specialized Persona Passes
Progress marker: Before beginning Pass 2, emit:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Pass 2 — Specialized Persona Passes
Personas: P1 Reliability Engineer | P2 Concurrency Specialist | P3 API Contract Reviewer
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Load $PLUGIN_DIR/skills/code-review/references/pass2-personas.md for persona definitions.
Run three focused reviews sequentially. Each persona:
- Receives the file set + ALL prior findings (Pass 1 + prior personas)
- Has a de-duplication gate (MUST NOT re-report same file + defect_class)
- Must cite file/function/evidence
- Max 5 findings each
| Order | Persona | Lens | Activation |
|---|---|---|---|
| P1 | Reliability Engineer | Error propagation, failure modes, recovery | Always |
| P2 | Concurrency Specialist | Cross-component races, shared state, async pitfalls | Only if concurrency signals detected |
| P3 | API Contract Reviewer | Interface misuse, boundary validation, contract mismatch | Always |
De-duplication instruction (inject before each persona)
You have already found these findings in prior passes:
{summary: CID/fingerprint, file, checker/defect_class for each}
DO NOT re-report any finding covering the same file + defect class.
Only report NEW findings that add information not captured above.
Pass 2 output
Findings follow the format in $PLUGIN_DIR/skills/code-review/references/pass2-personas.md. Fingerprint
fixable findings per $PLUGIN_DIR/skills/shared/fingerprint-spec.md. Set Pass: 2.
Progress marker: After all active Pass 2 personas complete, emit:
✅ Pass 2 complete — {N} additional finding(s) across {active_count} persona(s)
(Count only personas that ran — exclude any that were skipped.)
Pass 3 — Free-Flow Adversarial Pass
Progress marker: Before beginning Pass 3, emit:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Pass 3 — Free-Flow Adversarial Pass
De-scoped from all prior findings. Architectural + interaction risks.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
De-scoped from everything above. Receives ALL prior findings and MUST NOT re-report any of them.
Instruction
Analyze the code and all prior findings. Focus on what deterministic checkers
and persona reviews typically miss:
- Architectural smell and hidden coupling between components
- Interaction effects that emerge from how components compose
- Design patterns that will break when requirements change
- Subtle correctness issues in business logic
- Performance traps that only manifest under load
- "What will break when this code is modified next?"
Report the most impactful risks (up to 7). Only include risks you can tie to
specific evidence — cite the file, function, or pattern observed.
For each risk:
- Name (plain language)
- Impact (one line with justification)
- Why it matters in THIS codebase (reference specific code)
- A specific mitigation (code change, not generic advice)
Label the section: "LLM-inferred code quality hypotheses — validate before
treating as confirmed defects."
Pass 3 output
Advisory findings. Not fingerprinted. Not entered in the ledger unless promoted. Set Pass: 3.
Progress marker: After Pass 3 completes, emit:
✅ Pass 3 complete — {N} risk hypothesis/hypotheses (advisory, not in ledger)
(Use "hypothesis" for N=1, "hypotheses" for N≠1.)
Post-Scan — Report Assembly
HTML report
Generate a self-contained HTML report with:
- Summary — CID count by severity, scan metadata
- Pass 1 Findings — Coverity-style findings table with event paths
- Pass 2 Findings — "Expert Analysis" section, by persona
- Pass 3 Findings — "Code Quality Hypotheses" section (labeled advisory)
- Summary Table — CID, Checker, Impact, File, Function, Status
Write to CodeReviews/code-review-{YYYY-MM-DD}.html and Markdown report.
Ledger update
Write or update CodeReviews/code-review-ledger.md per $PLUGIN_DIR/skills/shared/ledger-schema.md.
Read existing ledger first:
cat CodeReviews/code-review-ledger.md 2>/dev/null || echo "NO_LEDGER_YET"
Apply reconciliation rules. Create folder and add to .gitignore:
mkdir -p CodeReviews
grep -q "^CodeReviews/" .gitignore 2>/dev/null || echo "CodeReviews/" >> .gitignore
Cache update
Update .claude/file-cache.json per $PLUGIN_DIR/skills/shared/file-cache-schema.md:
- Update charCount, lastScanned, add "code-review" to scannedBy
- Merge with existing entries
Checkpoint cleanup
Delete .claude/code-review-checkpoint.json on successful completion.
Reference Files
| File | When to load |
|---|---|
$PLUGIN_DIR/skills/code-review/references/checkers.md | Always — universal checker categories |
$PLUGIN_DIR/skills/code-review/references/checkers-dotnet.md | C# / ASP.NET / WCF / EF / Dapper |
$PLUGIN_DIR/skills/code-review/references/checkers-typescript.md | Angular / TypeScript / Node.js |
$PLUGIN_DIR/skills/code-review/references/checkers-java.md | Java / Spring Boot / JPA |
$PLUGIN_DIR/skills/code-review/references/checkers-python.md | Python / FastAPI / Django / Flask |
$PLUGIN_DIR/skills/code-review/references/pass2-personas.md | Always — Pass 2 persona definitions |
$PLUGIN_DIR/skills/code-review/references/analysis-rules.md | Always — analysis meta-rules |
$PLUGIN_DIR/skills/code-review/references/output-format.md | Always — finding format |
$PLUGIN_DIR/skills/code-review/references/webconfig-checks.md | ASP.NET Framework web.config analysis |
$PLUGIN_DIR/skills/shared/three-pass-spec.md | Architecture reference |
$PLUGIN_DIR/skills/shared/interactive-menu-spec.md | Menu specification |
$PLUGIN_DIR/skills/shared/scope-flags-spec.md | Scope flag definitions |
$PLUGIN_DIR/skills/shared/file-cache-schema.md | Cache schema and merge rules |
$PLUGIN_DIR/skills/shared/checkpoint-schema.md | Checkpoint schema |
$PLUGIN_DIR/skills/shared/fingerprint-spec.md | Fingerprint generation |
$PLUGIN_DIR/skills/shared/ledger-schema.md | Ledger format and reconciliation |
$PLUGIN_DIR/skills/shared/business-context-severity.md | B1-B7 override triggers |
$PLUGIN_DIR/skills/shared/source-file-consent.md | Consent category enforcement |
$PLUGIN_DIR/skills/shared/dismissed-findings-reconciliation.md | Rule 5 dismissed finding handling |
$PLUGIN_DIR/skills/shared/graph-index-schema.md / graph-module-schema.md | Knowledge graph for --area |