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 the bash interpreter. This pattern allows the execution of arbitrary code not reviewed in the skill's source.
[COMMAND_EXECUTION]: User-provided input (the application "idea") is assigned to a shell variable and then used in a command substitution. This creates a potential command injection surface where malicious input could escape the variable assignment and execute arbitrary shell commands.
[PROMPT_INJECTION]: The skill fetches a dynamic Markdown "runbook" from a remote URL (https://basic-anteater-667.convex.site/q/<SLUG>.md) and directs the agent to follow its instructions as the "complete, current rule set." This creates a surface for indirect prompt injection.
Ingestion points: Remote Markdown content fetched from the Convex platform (SKILL.md).
Boundary markers: None; the agent is explicitly instructed to follow the fetched rules over the skill's own summary.
Capability inventory: Full bash tool access, background process execution, and local filesystem monitoring.
Sanitization: No sanitization or validation of the remote instruction content is performed before processing.
[DATA_EXFILTRATION]: The skill transmits the user's project description to an external server at https://basic-anteater-667.convex.site/generate via a POST request for telemetry and project generation.

quickstart