input-validation-audit
Input validation audit: statically detects all input surfaces (HTTP params, CLI args, env vars, file reads, stdin) across a codebase, classifies their validation state (none/weak/adequate), and flags
SKILL.mdskills/security/input-validation-audit/SKILL.md
GitHubWhat this is for
Every external input crosses a trust boundary. HTTP request parameters, CLI arguments, environment variables, file contents, and stdin are untrusted until validated. In practice, validation is often missing, inconsistent, or applied too late.
This skill finds every input surface in a codebase and checks whether validation exists nearby. It combines a deterministic collector (regex-based detection of 8 surface types across 10 languages) with LLM context analysis that validates each finding, filters false positives, and produces an actionable report.
Audience: Senior > Vibe
- Seniors use it as a systematic pre-release validation gap audit and a way to prove that all external inputs are accounted for.
- Vibe-coders get an automatic safety net: the LLM explains what's at risk and how to fix it, so every finding is a learning opportunity.
Trigger: /input-audit
Intervention-free static analysis. The collector reads only - never writes, never executes the target code, never connects to a network.
PROTECTION RULE - never ~/.claude/
This skill analyzes foreign projects. It is read-only. The protection guard is still active: if the skill ever gets a write mode, target path validation must be implemented.
What You Must Do When Invoked
Step 1 - -help/-h check
If invoked with -help or -h, print the usage block below and stop.
Step 2 - Confirm -ProjectDir
If not provided, prompt the user. Confirm the path exists. Print:
Input validation audit on <path> ...
Step 3 - Run Collector
& .\scripts\input-scan.ps1 -ProjectDir "<path>"
Parse the JSON output. If exit code != 0, report error and stop.
Step 4 - Validate Each Finding
For each finding in the JSON:
- Read the
contextblock (5 lines around the surface line). - Determine if this is a true positive, false positive, or uncertain.
- Assign confidence:
proven,likely,suspected. - For
env-varsurfaces with hardcoded fallback values: check if the fallback constitutes validation (e.g.,port = process.env.PORT || 3000is not real validation - it only provides a default, not a format check). - For
sqlsink surfaces: confirm the SQL query is indeed parameterized or concatenated with the input.
Step 5 - Consistency Analysis
Group findings by parameter name across different files. Flag when:
- Same field (e.g.,
email,userId) validated in one handler but not another - Different validation approaches for semantically equivalent fields
Step 6 - Produce Report
Write input-validation-report.md to the working directory:
# Input Validation Report - <project-name>
## 1. Summary
- <count> input surfaces: <n> unvalidated, <m> validated
- <p> high-severity, <q> medium-severity, <r> low-severity
## 2. Critical Gaps (high severity + unvalidated)
| # | File | Line | Surface | Sink | Context |
|---|---|---|---|---|---|
| 0 | src/api/users.ts | 42 | req.body.email | sql | ... |
## 3. Medium Gaps
Grouped by surface type: http-query, http-body, env-var, etc.
## 4. Low / Informational
Validated surfaces with risky patterns (env fallback, type coercion only).
## 5. Consistency Issues
Same field name, different validation across handlers.
## 6. Validation Patterns Catalog
Libraries and approaches found: Joi, Zod, typeof checks, etc.
## 7. False Positives Removed
Each dismissed finding with reason.
## 8. Open Questions
Uncertain findings needing manual review.
Usage
# Full scan
& .\scripts\input-scan.ps1 -ProjectDir "C:\Projects\my-api"
# Scan with severity threshold
& .\scripts\input-scan.ps1 -ProjectDir "C:\Projects\my-api" -MinSeverity high
# Scan specific extensions only
& .\scripts\input-scan.ps1 -ProjectDir "C:\Projects\my-api" -Extensions "*.js,*.ts"
# Exclude test files
& .\scripts\input-scan.ps1 -ProjectDir "C:\Projects\my-api" -Exclude "test,spec,fixture"