The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses prescriptive and high-pressure language to override the agent's standard operational guidelines. Instructions such as "The script IS the workflow," "Deviating from the script HARMS the user," and "Do NOT explore or detect first" are designed to bypass the agent's autonomous verification and exploration steps.
[COMMAND_EXECUTION]: The skill's primary function is the execution of a Python module (python3 -m skills.incoherence.incoherence) with arguments that control its state machine.
[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that passes accumulated state and context via the --thoughts argument. This creates a feedback loop where data processed by the external script is re-injected into the agent's context and then passed back to the script, representing a significant injection surface.
Ingestion points: The --thoughts argument in SKILL.md accepts state derived from previous agent/script interactions.
Boundary markers: None identified in the prompt templates or instructions to distinguish between trusted instructions and untrusted data within the thoughts context.
Capability inventory: The skill has the ability to execute Python scripts via the shell.
Sanitization: There is no evidence of sanitization or filtering of the content passed through the <context> placeholder in the --thoughts argument.

incoherence