The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses and modifies sensitive configuration files located at ~/.claude/settings.json, .claude/settings.json, and .claude/settings.local.json. These files define the agent's behavior and execution environment. While this access is core to the style management functionality, these paths represent high-privilege configuration data.
[COMMAND_EXECUTION]: The instructions explicitly direct the agent to execute shell commands using jq or python3 to perform structured edits on JSON configuration files.
[PROMPT_INJECTION]: The skill ingests data from user-created markdown files in ~/.claude/output-styles/ to define agent output behavior, creating a surface for indirect prompt injection.
Ingestion points: Any *.md file located in ~/.claude/output-styles/ or .claude/output-styles/.
Boundary markers: Not present; the instructions do not include markers or warnings to isolate ingested style content.
Capability inventory: Local file system read/write, JSON modification via jq and python3, and markdown file creation.
Sanitization: The use of jq and python3 for JSON manipulation provides structural validation, but no content sanitization is specified for the instructional text within the styles.

style