connect
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- Indirect Prompt Injection (HIGH): The skill is explicitly designed to read and process data from over 1000 external applications (e.g., GitHub issues, Slack messages, emails) which are inherently untrusted sources.
- Ingestion points: Data enters the agent context via integrations with services like Gmail, Slack, and GitHub (referenced in 'Supported Apps' and 'Examples' sections).
- Boundary markers: None detected. The skill instructions do not provide delimiters or warnings to ignore instructions embedded within the data retrieved from these apps.
- Capability inventory: The skill has extensive high-privilege capabilities including sending emails, creating GitHub issues, posting to Slack, and updating databases (mapped to 'allowed-tools: Bash, Read, Write, Edit').
- Sanitization: No evidence of sanitization or validation of external content before it is processed by the agent.
- Risk: An attacker could send an email or post a Slack message containing hidden instructions that the agent would then execute, such as 'Delete all files' or 'Exfiltrate database content'.
- External Downloads (LOW): The skill requires installing several external packages.
- Evidence:
pip install composio,npm install @composio/core,pip install claude-agent-sdk,pip install composio-langchain. - Status: These are recognized packages, and under [TRUST-SCOPE-RULE], these findings are downgraded to LOW, but they remain necessary for the skill's functionality.
- Command Execution (MEDIUM): The skill documentation encourages the use of
exportfor setting environment variables andpip/npmfor installation. - Evidence:
export COMPOSIO_API_KEY="your-key",pip install composio. - Risk: While these are setup instructions, they involve shell execution that could be exploited if modified in a malicious version of the skill.
- Data Exfiltration (LOW): The skill's primary function is to send data to external APIs.
- Evidence: Network operations are performed via the Composio Tool Router to various third-party domains (Gmail, Slack, etc.).
- Status: While this is the intended purpose, it facilitates the movement of data out of the local environment to third-party services.
Recommendations
- AI detected serious security threats
Audit Metadata