The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from the repository (Read/Grep/Glob/Bash) and external documentation (Agent tool) to build planning artifacts. This creates a surface where poisoned content could influence execution plans. 1. Ingestion points: Local files (Bash/Read/Grep/Glob) and external documentation (Agent tool). 2. Boundary markers: Uses structured voices (Metis/Momus/Oracle) and a 'clean-room' design; lacks explicit data delimiters. 3. Capability inventory: File system access (Bash), downstream tool invocation (ultragoal, team), and internal state writing (.omw/). 4. Sanitization: Relies on a multi-pass adversarial scan (Momus) and machine-checkable self-verification (Oracle) to validate the resulting plan.\n- [EXTERNAL_DOWNLOADS]: The skill utilizes the Agent tool for research fan-out to gather Open Source Software (OSS) evidence and official documentation, which is a standard part of the planning process.\n- [COMMAND_EXECUTION]: The skill uses Bash for 'read-only repository inspection' to verify referenced files, commands, and project conventions without modifying the source code.

prometheus-strict