The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (scripts/generate-content.py) to process files and build prompts. This script is intended for local automation but accepts arbitrary file paths as arguments.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it ingests data from external source files and interpolates it directly into the LLM prompt without sufficient isolation.
Ingestion points: The scripts/generate-content.py script reads markdown files from the filesystem (Obsidian vault or custom paths).
Boundary markers: Absent. The script joins source content using simple triple-dash (---) separators, which do not protect against embedded instructions.
Capability inventory: The skill can read files, write files to the output directory, and execute shell commands.
Sanitization: No sanitization or instruction-filtering is applied to the data retrieved from source files.
[DATA_EXFILTRATION]: The script scripts/generate-content.py has the capability to read any file on the local filesystem if provided with an absolute path via the --sources argument. Additionally, it explicitly attempts to load environment variables from the agent's internal configuration directories (~/.claude/.env), which could lead to the exposure of sensitive credentials if the agent includes that content in its prompt output.

cmo-content-generator