ctx

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). The required runtime workflow includes calling ctx-cli mcp call get_cve_resolution_status, whose response embeds data.recommendedAction as free-form diff/advisory text; that text originates from the tenant’s CVE resolution pipeline (i.e., non-user-authored security advisories/diffs) and is ingested into the agent/LLM context via the tool output.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill contains runtime curl commands that fetch and install an executable from GitHub releases (e.g. https://github.com/tabnine/skills/releases/download/$TAG/ctx-cli-...), which downloads remote code that would be executed on the host.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill includes explicit install commands that download a binary to /usr/local/bin and run chmod (modifying system-level files that typically require elevated privileges), which instructs changing the machine's state and could be used to compromise it.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)