The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a --force flag for the --next command, which is explicitly documented to "bypass safety gates." This provides a mechanism for the agent to override or ignore established safety constraints and validation steps during workflow execution.
[INDIRECT_PROMPT_INJECTION]: The --do flag accepts freeform natural language intent strings via $ARGUMENTS and dispatches them to other workflows. This creates a vulnerability surface where malicious instructions embedded in the input (e.g., if the input is sourced from an untrusted file or external tool) could influence the agent's subsequent actions.
Ingestion points: User-provided $ARGUMENTS processed in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the freeform input string.
Capability inventory: The skill is authorized to use powerful tools including Bash, Read, Grep, Glob, and SlashCommand.
Sanitization: There is no evidence of input sanitization or validation before the intent is dispatched to the routing logic.

gsd-progress