The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides specific command-line examples for the agent to run, including the use of tools like ysoserial, phpggc, and ysoserial.net. These commands are used to generate exploit payloads and perform out-of-band (OOB) network requests using curl and nslookup to verify code execution.
[DYNAMIC_EXECUTION]: The cheatsheet.md file contains a Python script template that uses the pickle module and the __reduce__ magic method to generate a serialized object capable of executing arbitrary shell commands. The agent is encouraged to execute this script to generate payloads for testing.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze potentially malicious or attacker-controlled data from external web application components such as cookies, hidden fields, and parameters. This introduces a vulnerability surface where the target application's data could influence the agent's behavior.
Ingestion points: Data retrieved from the target application, including HTTP cookies, parameters, hidden form fields, and message queues as described in SKILL.md.
Boundary markers: None. The instructions do not provide delimiters or specific guidance on isolating untrusted data from the agent's operational instructions.
Capability inventory: The agent is granted capabilities to execute shell commands, run Python scripts, and perform network operations (curl, nslookup).
Sanitization: No sanitization, escaping, or validation techniques are mentioned for handling the data retrieved from external targets.

web-deserialization