The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a mechanism to ingest, process, and persist user-defined templates which are later retrieved and used to structure the output of subsequent agent interactions.
Ingestion points: The skill takes "Template content" as direct input from the user in SKILL.md.
Boundary markers: There are no defined delimiters or safety instructions specified to isolate the user-provided template content, increasing the risk that the agent may follow instructions embedded within the template rather than treating them as data.
Capability inventory: The skill has the capability to write directly to the file system (~/.claude-marketing/brands/{slug}/templates/) and execute a management script (scripts/guidelines-manager.py).
Sanitization: No validation or sanitization step is described to ensure that the template content does not contain prompt injection payloads meant to override agent behavior during the formatting phase of other commands.
[COMMAND_EXECUTION]: The skill executes a local script to perform CRUD operations on the templates.
Evidence: The process uses scripts/guidelines-manager.py to save data to the file system. While this is a local utility, the use of user-supplied variables like {slug} and {name} in the command line (guidelines-manager.py --brand {slug} --action save-template --name {name}) presents a potential surface for command injection if these variables are not strictly validated.

import-template