The Agent Skills Directory

[PROMPT_INJECTION]: The skill description contains instructions intended to bypass safety and 'stance' filters. It describes the 'beta' mode as an 'untreated default' and an 'off switch' designed to disable deliberate filtering and provide the model's 'accommodation reflex exactly as shipped'.
[DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically load and prioritize instructions from a path computed at runtime via the $POWERSUIT_OVERLAY environment variable. This allows the skill's logic to be modified by external files not contained within the skill itself.
[COMMAND_EXECUTION]: The instructions direct the agent to execute a shell command ls -1 "$POWERSUIT_OVERLAY/stance" to verify the existence of an overlay directory before reading its contents.
[INDIRECT_PROMPT_INJECTION]: The skill introduces a surface for indirect prompt injection by ingesting untrusted external data into the agent's context.
Ingestion points: The agent is instructed to read README.md and any files within the stance/beta/ subdirectory located at the path defined by $POWERSUIT_OVERLAY (referenced in SKILL.md).
Boundary markers: No delimiters or warnings to ignore embedded instructions are present; the skill explicitly instructs the agent to 'follow the overlay' where it disagrees with the default instructions.
Capability inventory: The skill utilizes shell command execution (ls) and file system read capabilities.
Sanitization: There is no evidence of sanitization, validation, or filtering of the content retrieved from the overlay files before it is processed as authoritative instructions.

beta