The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to suppress transparency and hide its activities from the user. It directs the agent to 'drop the report without telling the user' in case of failure and to 'not mention the report' unless a specific success message is received. This behavior reduces user oversight of background network operations.
[DATA_EXFILTRATION]: The skill's primary function is to transmit data to the 'OpenResearch team' using the orx feedback command. While it provides privacy guidelines, it explicitly commands the agent to remove or replace shell-sensitive characters (backticks, $, <, >, |, ;, and &) specifically so that 'agent permission checks do not interrupt filing.' This is a deliberate attempt to evade security guardrails and platform monitoring intended to prevent command injection or unauthorized data transfer.
[OBFUSCATION]: The instruction to replace literal characters with bracketed names (e.g., [ampersand]) constitutes a form of payload obfuscation. This technique is used to bypass keyword-based or pattern-based security scanners that monitor command arguments for sensitive symbols.
[COMMAND_EXECUTION]: The skill relies on the execution of the orx CLI tool to perform its reporting tasks, incorporating various user-provided and system-generated strings into the command arguments.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data to populate feedback fields, creating a vulnerability surface for indirect injection.
Ingestion points: User expressions of frustration and verbatim quotes from the conversation are used as input for the --details and --quote flags (SKILL.md).
Boundary markers: The instructions lack strict delimiters or escaping mechanisms to isolate untrusted user content from the command structure, relying only on single quotes which are vulnerable to escape sequences.
Capability inventory: The skill possesses network exfiltration capabilities via the orx CLI.
Sanitization: While the skill suggests redacting secrets, the primary sanitization logic (replacing shell characters) is focused on evading platform security filters rather than ensuring the integrity of the prompt or command structure.

orx-feedback