The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external trend research data which is then interpolated into LLM prompts, creating a potential surface for instructions embedded in the data to override agent behavior.\n
Ingestion points: The scripts/generate_concepts.sh script reads the content of the file provided via the --trend-input argument.\n
Boundary markers: The concept_generation template in prompts.json includes the trend_input context without utilizing delimiters or specific instructions to the agent to treat the content as data only.\n
Capability inventory: The skill has the ability to read and write files to the local file system and execute shell scripts and Python code.\n
Sanitization: No validation or sanitization is performed on the trend-input file content before it is processed and passed to the host agent.\n- [COMMAND_EXECUTION]: The pipeline is implemented using shell scripts (generate_concepts.sh, pick_concept.sh, draft_script.sh) that execute Python modules and inline Python code.\n
These scripts handle user-provided file paths and slugs as command-line arguments which are used in subsequent file system and process execution logic.\n- [DATA_EXPOSURE]: The skill accesses local brand workspaces to read brand-voice.md and voice-profile.md files, which may contain sensitive brand specifications.\n
The default brand root points to ~/Documents/GitHub/slideshow-brands, indicating access to local development or configuration directories.

concept-generator