Security Scan Skill

Security Scan Skill

Scan the codebase for security issues across three categories:

1. Dependency vulnerabilities — Known CVEs in packages
2. Static analysis — Insecure code patterns (per project tooling)
3. Secrets detection — API keys, passwords, tokens in code

When This Skill Runs

• Automatically during /​phase-checkpoint
• On-demand via /​security-scan
• Optionally during /​verify-task for security-critical tasks

Workflow Overview

Copy this checklist and track progress:

Security Scan Progress:
- [ ] Step 1: Discover security tooling from project docs
- [ ] Step 2: Run dependency audit
- [ ] Step 3: Run secrets detection
- [ ] Step 4: Run static analysis
- [ ] Step 5: Aggregate and deduplicate findings
- [ ] Step 6: Present issues with severity
- [ ] Step 7: Offer to apply fixes

Step 1: Discover Project Security Tooling

Read project documentation and task runners to find the correct commands:

• README.md
• CONTRIBUTING.md
• SECURITY.md
• Makefile
• Taskfile.yml
• justfile
• Any security or build scripts under scripts/

Extract any documented commands for:

• Dependency auditing
• Static analysis / security scanning
• Secrets detection (if the project has a preferred tool)

If nothing is documented, ask the human to provide the correct commands.

Step 2: Dependency Audit

• If a dependency audit command is documented or provided, run it.
• If no command is available, mark this check as SKIPPED and note it in the report.

Step 3: Secrets Detection (Default)

Run a pattern-based secrets scan (stack-agnostic) unless the project documents its own secrets tool. If a project-specific tool exists, use it instead.

Patterns to Detect

Directories to Skip

• node_modules/
• .git/
• vendor/
• venv/, .venv/, env/
• dist/, build/
• *.min.js, *.bundle.js
• Binary files

Step 4: Static Analysis

• If a static analysis or security scanning command is documented or provided, run it.
• If no command is available, mark this check as SKIPPED and note it in the report.

Step 5: Aggregate Findings

Collect all findings into a unified format:

SECURITY SCAN RESULTS
=====================

Scanned: {timestamp}
Checks Run: Dependencies | Secrets | Static Analysis

CRITICAL (N)
------------
[issue details]

HIGH (N)
--------
[issue details]

MEDIUM (N)
----------
[issue details]

LOW (N)
-------
[issue details]

Summary: {N} critical, {N} high, {N} medium, {N} low

Step 6: Present Issues

For CRITICAL and HIGH issues, present interactively with resolution options. If a fix command is documented, offer it as the primary option.

Step 7: Propose Fixes

Propose fixes to the user. Do not apply fixes automatically — present each fix for user approval before making changes.

For each fix, show a preview:

• Display the proposed change (file, line, before/​after)
• Ask for confirmation: "Apply this fix?" (Yes/​Skip)
• Only modify files after explicit user confirmation

Apply fixes based on user choices:

• Use project-documented fix commands when available
• Otherwise, propose manual code changes and confirm before editing

Output Format

Structured result (for all callers)

When invoked by any skill (/​phase-checkpoint, /​verify-task, or others), return:

Security Scan: PASSED | FAILED | PASSED WITH NOTES

Issues: X critical, Y high, Z medium
Fixed: N issues
Skipped: M checks (documented)

Blocking: Yes/​No

This format is the same regardless of the caller. The calling skill decides how to interpret it (e.g., /​phase-checkpoint may block on FAILED; /​verify-task may log it as a note).

When called standalone

Show the structured result above followed by the full report with all findings and interactive fix options (Step 6 and Step 7).

Severity Definitions

Tool Installation Notes

If required tools are missing, instruct the user to install them based on the project's documentation or security policy.

Error Handling

If no project documentation is found:

• Report: "No security tooling documentation found"
• Ask the user for the correct commands
• Proceed with default secrets detection (always available)

If a security tool returns a non-zero exit code:

• If output indicates findings (common for npm audit, pip-audit, etc.), treat as SUCCESS_WITH_FINDINGS
• Parse and report the findings; do NOT mark the check as FAILED
• If output indicates an execution error (crash, invalid args, config error), treat as FAILED
• Always continue with other checks and include the outcome in the final report

If tool is not installed:

• Report: "{tool} not found"
• Provide installation instructions if known
• Mark check as SKIPPED with reason
• Continue with available tools

If secrets scan finds too many results (>100):

• Truncate results with: "Showing first 100 of {N} findings"
• Suggest the user may have sensitive data that should be gitignored
• Recommend running on specific directories

If project has no package manager (dependency scan N/​A):

• Mark dependency scan as NOT APPLICABLE
• Note: "No package.json, requirements.txt, go.mod, or similar found"
• Proceed with other checks

If scan is interrupted:

• Report partial results obtained
• Mark interrupted checks clearly
• Suggest re-running the scan

Example Invocations

/​security-scan              # Full scan
/​security-scan --deps       # Dependencies only
/​security-scan --secrets    # Secrets detection only
/​security-scan --code       # Static analysis only
/​security-scan --fix        # Auto-fix where possible