Skip to content

Security: zhoudianfeng51-a11y/agent-command-center

Security

SECURITY.md

Security

Agent Command Center is local-first. The public demo is intentionally narrow and does not call real executors.

Demo-Only Defaults

The default configuration keeps real execution off:

ENABLE_REAL_EXECUTION=false
ENABLE_OPENCLAW=false
ENABLE_QCLAW=false
ENABLE_CODEX=false
ENABLE_FILE_UPLOAD=false
ENABLE_DAILY_SCAN=false

Without explicit external-agent configuration, the app will not call OpenClaw, QClaw, Codex, GPT, or other external tools automatically.

File Access Boundary

The demo server only previews relative files under examples/sample-data/.

Future local file access must use an allowlist such as:

COMMAND_CENTER_ALLOWED_ROOTS=./examples/sample-data,./data

Do not add unrestricted absolute-path reads. Do not follow paths outside configured roots.

Data That Should Not Be Committed

Do not commit:

  • real data/ runtime records;
  • real logs;
  • real receipts;
  • real task cards;
  • uploaded files;
  • local machine paths;
  • credentials, tokens, private keys, or environment files.

The repository .gitignore excludes data/, dist/, node_modules/, build cache files, and local environment files.

Reporting Security Issues

Until a private disclosure channel is added, open a GitHub issue with a minimal description that does not include secrets, private files, or exploit payloads. If the issue involves sensitive information, describe the class of problem and ask maintainers to arrange a private channel before sharing details.

Before publishing a fork, run:

npm run smoke
npm run build

There aren't any published security advisories