The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool in Step 9 to execute directory creation commands: mkdir -p "$(dirname "${output_path}")" and mkdir -p "{source_dir}/cogni-visual". The output_path and source_path (from which source_dir is derived) are parameters that can be supplied by the user. The instructions do not include any sanitization or validation logic to handle shell metacharacters, creating a direct path for command injection if a malicious path string is provided.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes external narrative documents.
Ingestion points: Narrative files are auto-discovered or provided via source_path and read during Step 0 and Step 2.
Boundary markers: The instructions lack the use of delimiters or specific 'ignore' directives when processing the narrative content, allowing instructions embedded in the data to potentially override the skill's logic.
Capability inventory: The skill has broad capabilities including file writing (Write, Edit), file system navigation (Glob, Grep), and shell execution (Bash).
Sanitization: There is no evidence of content filtering, escaping, or validation of the narrative data before it is processed by the agent.

story-to-storyboard