The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external markdown files (CONTEXT-*.md, tasks.md, manifest.yaml) and treats instructions found within the ## Standing section as active constraints and rules for the agent to follow. This creates a surface where a malicious or compromised context file could attempt to override agent behavior.
Ingestion points: CONTEXT-{stream}-llm.md, OpenSpec project/proposal/tasks.md, manifest.yaml, and $PRAXIS_DIR/thinking artifacts.
Boundary markers: Files are read directly into the context; instructions note that the Phase 1 read puts the 'whole file in context'.
Capability inventory: Uses Bash (via rtk), Read, and AskUserQuestion tools.
Sanitization: No explicit sanitization or filtering is performed on the content of the ## Standing section before it is adopted as active rules.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute rtk ls commands for file discovery and session listing. While limited to directory listing in the provided instructions, it represents shell-level access to the file system.

load-context