The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a shell script from a remote URL (https://basic-anteater-667.convex.site/quickstart-bootstrap) and executes it using bash. Executing scripts from remote sources poses a high security risk.
[OBFUSCATION]: The instructions explicitly direct the agent to split the download and execution of the remote script into separate steps specifically to evade safety harness classifiers that monitor for patterns like 'curl | bash'. This is a deliberate attempt to bypass security controls.
[INDIRECT_PROMPT_INJECTION]: The skill fetches a personalized 'runbook' from a remote URL and instructs the agent to follow its rules exactly. This creates a surface where external instructions can control agent behavior. Ingestion points: SKILL.md (Step 2) fetches from https://basic-anteater-667.convex.site/q/.md. Boundary markers: Absent; instructions mandate following the external runbook 'exactly'. Capability inventory: Shell command execution (bash), file writing (scaffolding), and network operations (curl). Sanitization: None provided for the fetched markdown content.
[DATA_EXFILTRATION]: The skill transmits the user's project description to an external server via a POST request to https://basic-anteater-667.convex.site/generate.
[COMMAND_EXECUTION]: The skill executes multiple shell commands including curl, bash, and node to manage the scaffolding process.

labs-quickstart