The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs coercive and forceful rhetoric designed to override the agent's standard behavioral and safety boundaries. It explicitly instructs the agent to disregard its own assessments of task feasibility (e.g., "forbidden from saying 'I cannot solve this'") and uses high-pressure role-play ("Western big-tech performance culture") to compel the AI to bypass its default operational constraints.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a high-risk surface for indirect prompt injection by requiring the agent to ingest and analyze large volumes of untrusted data without safety boundaries.
Ingestion points: The agent is mandated to read 50+ lines of local source code, configuration files, and external search results "word by word" during the "Elevate" and "Audit" phases.
Boundary markers: The instructions lack any requirement for delimiters or "ignore embedded instructions" warnings when processing this external content.
Capability inventory: The skill utilizes powerful tools including bash (shell execution), view_file (file reading), and edit_file (filesystem writes), which could be misused if the agent follows instructions embedded in the data it is auditing.
Sanitization: No sanitization or escaping of external content is required before the agent interpolates that data into its internal reasoning or next-step execution.
[DYNAMIC_EXECUTION]: The "Red-Capable Loop" and "Debug Discipline" protocols instruct the agent to dynamically generate and execute code at runtime, such as test harnesses, CLI scripts, and bisection tools. While intended for debugging, this involves the runtime execution of agent-generated code that may be influenced by untrusted data processed during the audit phase.
[COMMAND_EXECUTION]: The skill encourages the frequent and proactive use of shell commands (bash) to "exhaust all options" and "verify underlying assumptions." This increases the potential impact of command injection if the agent is manipulated by malicious data in the environment or files it is inspecting.

pua