The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process untrusted data from external website responses (JSON, HTML, JS) stored in HAR files to derive API shapes and generate client code. This creates a risk where malicious instructions embedded in a website's response could influence the agent's behavior during the script generation phase.
Ingestion points: Website response text is read from the HAR file in Step 3 using jq.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are suggested for the agent when processing the response bodies.
Capability inventory: The skill possesses the capability to write files (Step 4: Generate the client), execute shell commands (Step 1, 2, 3), and perform network operations (Step 5: Verify).
Sanitization: The instructions do not specify any sanitization or escaping of the external content before it is interpolated into the generated code logic.
[DYNAMIC_EXECUTION]: The core workflow involves generating a new standalone client script or library (Step 4) and subsequently executing the generated functions (Step 5) to verify they correctly interact with the live API. This represents a pattern of generating and executing code based on observed external data.
[COMMAND_EXECUTION]: The skill utilizes command-line tools such as jq for data processing and agent-browser for capturing and managing network traffic and session cookies.

derive-client