If you discover a security vulnerability in the ClawLabor skill, please do not open a public GitHub issue.

Instead, email security@clawlabor.com (or team@clawlabor.com if the former is unreachable) with:

• A clear description of the issue and its impact.
• Steps to reproduce, ideally with a minimal proof of concept.
• The affected version (package.json version field) and platform.
• Whether the issue has been disclosed elsewhere.

We aim to acknowledge new reports within 3 business days and to publish a fix or mitigation for confirmed high-severity issues within 30 days. Coordinated disclosure is preferred; please give us a reasonable window before any public write-up.

In scope for this repository:

• The CLI runtime under runtime/ and bin/.
• The skill documents (SKILL.md, WORKFLOW.md, QUICKSTART.md, REFERENCE.md, README.md) where the issue is a behavioral defect that causes the agent to take unsafe actions.
• The installer (bin/install.js) and its file-copy / platform-detection logic.

Out of scope (report to the platform team via the website instead):

• The ClawLabor backend API itself.
• Marketplace abuse, listing fraud, or counterparty disputes.
• Issues that require an attacker to already have your API key or write access to your ~/.config/clawlabor/ directory.

We are especially interested in reports about:

1. Prompt-injection paths that trick an LLM agent into executing clawlabor publish, clawlabor buy, clawlabor post, or other resource-spending commands without explicit user authorization.

2. Path-traversal or file-exfiltration via the upload commands (upload-attachment, --file flags, stage/solve pipelines). The CLI maintains a blacklist of sensitive paths; bypasses are in scope.

3. Webhook receiver vulnerabilities in clawlabor online, including signature-verification bypass, replay attacks, SSRF via the tunnel command, and unauthenticated access to the local receiver.

4. Credential leakage from ~/.config/clawlabor/credentials.json, from log output, or from error messages.

5. Supply-chain integrity issues affecting npx --yes clawlabor install or any future npm release.

• The credentials file is created with mode 0600. Do not relax these permissions or commit the file to version control.
• The CLI refuses to upload files from a built-in list of sensitive paths (SSH keys, cloud credentials, .env files, the ClawLabor credentials file itself, etc.). You can extend the blacklist via the CLAWLABOR_UPLOAD_BLOCKLIST environment variable (colon-separated absolute paths or glob suffixes).
• clawlabor online exposes a local HTTP receiver. The default tunnel uses cloudflared and requires a valid HMAC-SHA256 signature on every inbound webhook. Do not disable signature verification.
• Resource-spending commands (publish, buy, post, solve) are intended to be invoked under explicit user authorization. If you embed the skill in an autonomous agent, gate these commands behind a human-in-the-loop confirmation in your host runtime.