🛡️ secure-code

Security guardian that helps you ship faster. It blocks hard-coded secrets and destructive commands before they land, flags OWASP Top 10 patterns as code is written, audits your supply chain, and guides minimal, tested fixes.


Version	1.0.0
Format	Agent Plugins 1.0 + `com.github.copilot` extensions
Works in	VS Code · GitHub Copilot CLI · GitHub Copilot app
Requires	Node.js 18+ and `git` on `PATH`
Network access	None. Offline rules; secret values are always redacted

What's inside

Component	Name	Location	Portable?
Skill	`secure-code-review`	skills/secure-code-review/	✅ Agent Plugins standard
MCP server	`vuln-scout` (stdio)	mcp.json → servers/vuln-scout.mjs	✅ Agent Plugins standard
Custom agent	`security-guardian` (read-only)	com.github.copilot/agents/	Copilot clients
Hooks	`sessionStart`, `preToolUse`, `postToolUse`	com.github.copilot/hooks/hooks.json	Copilot clients
Slash command	`/security-scan`	com.github.copilot/commands/	Copilot clients
Automation template	Nightly security sweep	automations/	VS Code only

`vuln-scout` MCP tools (all read-only)

Tool	What it finds
`scan_secrets`	Cloud keys, GitHub/Slack/Stripe tokens, private keys, connection strings, JWTs, hard-coded passwords (values redacted)
`scan_insecure_patterns`	SQL injection, XSS, command injection, eval, unsafe deserialization, weak hashes, TLS bypass, permissive CORS, path traversal, open redirect, hard-coded JWT secrets, client-side auth, and more, each mapped to a CWE with a fix
`audit_dependencies`	Missing lockfiles, unbounded versions, non-registry sources, known-compromised npm packages, risky install scripts, unpinned GitHub Actions, `pull_request_target` misuse, and script injection in workflows
`scan_snippet`	Checks proposed code before it's written
`explain_cwe`	CWE name, OWASP category, impact, and fix

Hooks: the policy backstop

Event	Behavior (`SECURE_CODE_MODE=enforce`, the default)
`sessionStart`	Adds secure-coding guidance to the session context
`preToolUse`	Deny: writing high-confidence secrets, `curl … \| sh`, `rm -rf /`/`~`/``, `chmod 777`, disk wipes, disabling TLS verification, reading SSH/cloud credential files, piping env vars to the network. Ask*: possible secrets, editing or reading `.env`/key files, `--no-verify`, `sudo`
`postToolUse`	Scans code that was just written and feeds medium+ findings (with CWE and fix) back to the agent so it corrects itself

These rules are fast, offline heuristics for the inner loop. They complement, and don't replace, GitHub code scanning (CodeQL), Dependabot, and secret scanning with push protection.

Install

copilot plugin marketplace add sekar3s/octocat-agent-plugins-demo
copilot plugin install secure-code@octocat-agent-plugins

VS Code and the GitHub Copilot app are covered in the installation guide.

Use it

Try this	Where
`/security-scan`	Any client
"Is this change safe to ship?" / "Review api/src/routes for vulnerabilities"	Any client (the skill loads automatically)
security-guardian agent	VS Code agent picker · CLI `copilot --agent secure-code:security-guardian` · Copilot app agent picker
`node skills/secure-code-review/scripts/scan.mjs --path <repo> --changed`	Terminal or CI (exits `1` on high/critical findings)

Suppress a reviewed finding

Add a comment containing secure-code-ignore on the finding's line or the line above it, with a reason:

// secure-code-ignore: identifiers come from a fixed allow-list (TABLES)
await db.run(`DELETE FROM ${table.name}`);

The scan script reports findings in test and fixture files but doesn't count them as blocking unless you pass --include-tests.

Configuration

Variable	Default	Effect
`SECURE_CODE_MODE`	`enforce`	`enforce` = deny + ask · `warn` = ask only (never deny) · `off` = disable hooks
`OCTOCAT_PLUGINS_AUDIT_DIR`	unset	When set, hook decisions are appended to `<dir>/secure-code.jsonl`. Commands are truncated; secret values are never logged

Remove

copilot plugin uninstall secure-code

See the uninstall guide for VS Code and the Copilot app.