The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest and explain "a body of work," code, or diffs. This creates a surface for indirect prompt injection where instructions embedded in the processed data could influence the agent's behavior.
Ingestion points: Processes external code, diffs, and debugging information as described in steps 2 and 5 of the instructions.
Boundary markers: Absent. There are no explicit instructions to use delimiters or to disregard instructions found within the data being analyzed.
Capability inventory: The skill is authorized to invoke other internal skills (how, why, unslop), read the filesystem, interact with a debugger, and call an image-generation tool.
Sanitization: Absent. Content is processed and reworded for teaching without filtering of potential embedded commands.
[COMMAND_EXECUTION]: The skill instructions direct the agent to invoke specific capabilities and tools to assist in its analysis.
Evidence: Steps 2 and 5 explicitly instruct the agent to "run how and why," "Open the diff, the code, or the debugger," and "reach for the image-generation tool."
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill uses a relative path to access configuration data from a source outside its immediate directory.
Evidence: The introductory block instructs the agent to "read ../orca-harness/SKILL.md" when the ORCA_CLI_COMMAND environment variable is set. While this appears to be for environment mapping, it involves navigating the file system to access sibling skill configurations.

teach