https://github.com/raalarcon9705/detect-project-malware
Agent skill that scans a project for supply-chain and infostealer patterns across JS/TS, Python, shell, configs and package.json. Works with Claude Code, Codex, Cursor, Gemini CLI, Antigravity and 40+ other agents.
https://github.com/raalarcon9705/detect-project-malware
agent-skill antigravity claude-code codex cursor gemini-cli incident-response infostealer malware-detection ripgrep security-scanner skill-md skills-sh static-analysis supply-chain-security
Last synced: 16 days ago
JSON representation
Agent skill that scans a project for supply-chain and infostealer patterns across JS/TS, Python, shell, configs and package.json. Works with Claude Code, Codex, Cursor, Gemini CLI, Antigravity and 40+ other agents.
- Host: GitHub
- URL: https://github.com/raalarcon9705/detect-project-malware
- Owner: raalarcon9705
- License: mit
- Created: 2026-04-21T19:26:27.000Z (2 months ago)
- Default Branch: main
- Last Pushed: 2026-04-21T19:36:14.000Z (2 months ago)
- Last Synced: 2026-04-21T21:35:37.207Z (2 months ago)
- Topics: agent-skill, antigravity, claude-code, codex, cursor, gemini-cli, incident-response, infostealer, malware-detection, ripgrep, security-scanner, skill-md, skills-sh, static-analysis, supply-chain-security
- Language: Shell
- Homepage:
- Size: 20.5 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Security: SECURITY.md
Awesome Lists containing this project
README
# detect-project-malware
An AI agent [Skill](https://skills.sh/) that audits an entire project for
supply-chain and infostealer patterns **without executing any of the suspect
code**. It works as a `SKILL.md` — compatible with Claude Code, Codex, Cursor,
OpenCode, Gemini CLI, Antigravity, and 40+ other agents that implement the
[Agent Skills open standard](https://github.com/vercel-labs/skills).
Built after a real incident where a contributor injected an obfuscated
credential-stealing payload into `postcss.config.js`, got it reverted, and
re-injected it weeks later behind an unrelated feature commit. The skill
automates the detection and the git-history attribution so the same class of
attack can be caught in seconds instead of days.
## What it catches
Covers JS/TS/JSX/TSX, Python, shell (bash/zsh/fish), Ruby, Go, PHP, Perl,
PowerShell, Dockerfiles, Makefiles, YAML, `package.json` install hooks, and all
JS/TS tooling configs. Groups hits into 14 categories:
- `JS_OBFUSCATION`, `JS_EVAL`, `JS_ESM_ESCAPE`, `JS_NATIVE_EXEC`
- `PY_EVAL`, `PY_NATIVE_EXEC`, `PY_REVERSE_SHELL`
- `SHELL_REMOTE_EXEC` — `curl | sh`, `bash <(curl …)`, `base64 -d | bash`, …
- `SHELL_REVERSE_SHELL` — `/dev/tcp/…`, `nc -e`, `bash -i >& /dev/tcp/…`
- `SHELL_PERSISTENCE` — `.bashrc` / `authorized_keys` / crontab / LaunchAgents / systemd
- `SHELL_DISABLE_SECURITY` — `setenforce 0`, `chattr -i`, iptables flush
- `CRED_BROWSER`, `CRED_SSH`, `CRED_CLOUD`, `CRED_TOKENS`, `CRED_WALLETS`,
`CRED_PRIVATE_KEYS`, `CRED_DISCORD`
- `NET_EXFIL` — Telegram bots, Discord/Slack webhooks, ngrok/transfer.sh, DNS beacons
- `MINING` — stratum, xmrig, coinhive, etc.
- `SUPPLY_CHAIN` — `postinstall` / `preinstall` / `prepare` hooks that fetch or exec
- `CONFIG_ANOMALY` — oversized single lines in config files (classic payload hiding spot)
See [PATTERNS.md](PATTERNS.md) for the full catalog with severity and typical
false-positive shapes.
## Install
### Via the `skills` CLI (Vercel Labs, works with 45+ agents)
```bash
npx skills add raalarcon9705/detect-project-malware
```
This auto-detects your installed agents (Claude Code, Codex, Cursor, OpenCode,
Gemini CLI, Antigravity, …) and places the skill in each agent's skills
directory. Install for a specific agent only:
```bash
npx skills add raalarcon9705/detect-project-malware --agent claude-code
npx skills add raalarcon9705/detect-project-malware --agent codex
npx skills add raalarcon9705/detect-project-malware --agent cursor
```
### Manual install
Clone the repo into the skills directory of your agent of choice:
| Agent | Path |
| -------------- | ------------------------------------------ |
| Claude Code | `~/.claude/skills/detect-project-malware` |
| Codex | `~/.codex/skills/detect-project-malware` |
| Cursor | `~/.cursor/skills/detect-project-malware` |
| Gemini CLI | `~/.gemini/skills/detect-project-malware` |
| Antigravity | `~/.antigravity/skills/detect-project-malware` |
| Generic | `~/.agents/skills/detect-project-malware` |
```bash
git clone https://github.com/raalarcon9705/detect-project-malware.git \
~/.claude/skills/detect-project-malware
```
## Usage
Once installed, tell your agent:
> Scan this repo for malware / supply-chain issues
Or invoke the scripts directly from any project root:
```bash
bash ~/.claude/skills/detect-project-malware/scripts/scan.sh
```
Typical output:
```
=== detect-project-malware scan ===
root: /Users/you/Development/your-repo
scanned: 713 file(s) in 2s
result: 3 suspicious file(s)
Categories seen:
CRED_TOKENS 2 match(es)
JS_ESM_ESCAPE 1 match(es)
● /Users/you/Development/your-repo/postcss.config.js
[JS_ESM_ESCAPE] createRequire
● /Users/you/Development/your-repo/test-goals.sh
[CRED_TOKENS] jwt-candidate
```
For any hit, attribute the injection commit:
```bash
bash ~/.claude/skills/detect-project-malware/scripts/find-injection.sh \
path/to/file.js --root .
```
Produces a CLEAN / INFECTED / INJECTED / CLEANED history walk plus the exact
`git show` command to restore the last clean version.
## Repository layout
```
detect-project-malware/
├── SKILL.md ← entry point (YAML frontmatter + workflow)
├── PATTERNS.md ← category catalog with severity & FP guidance
├── SECURITY.md ← threat model & how the scripts are safe to run
├── scripts/
│ ├── scan.sh ← ripgrep-based, multi-language scanner
│ ├── patterns.txt ← editable rule catalog (CATEGORY|name|regex)
│ └── find-injection.sh ← git-history attribution for a single file
├── LICENSE
└── README.md
```
## Requirements
- [ripgrep](https://github.com/BurntSushi/ripgrep) (`rg`) — the scanner uses it
for the fast file traversal and pattern matching.
- `bash` 3.2+ (scripts tested against macOS default bash and modern Linux).
- `git` 2.0+ (only for `find-injection.sh`).
## How it's safe to run
None of the scripts ever `source`, `eval`, `node`, `python`, `bash`, or
otherwise execute a suspect file. They only **read** content through ripgrep
and `awk`. Safe to point at a repo you already suspect is compromised.
See [SECURITY.md](SECURITY.md) for the threat model, scope, and what the skill
explicitly does **not** do.
## Contributing
Patterns live in `scripts/patterns.txt`. Format per line:
```
CATEGORY|name|regex
```
Lines starting with `#` and blank lines are skipped. Regex flavor: ripgrep
default (Rust regex, no look-around). When adding a new pattern, update
`PATTERNS.md` with a severity note and typical false-positive shape.
## License
[MIT](LICENSE) © 2026 Reynier Rivero