The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to manage local files, including directory creation (mkdir) and file listing (ls). It also executes local CLI utilities like maestro and internal scripts located in ~/.claude/get-shit-done/bin/ to initialize workflows and search for existing prompt patterns. These operations are restricted to the local environment and the vendor's own toolkit.
[DATA_EXFILTRATION]: The skill reads configuration files and existing prompt definitions from the project's .claude/ directory and the user's home directory. This information is used to extract patterns and inform the conversion process locally; no external network transfers were identified.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted data from existing files on disk. (1) Ingestion points: .claude/commands/, .claude/agents/, and .claude/skills/ (via SKILL.md steps 4a and 5c). (2) Boundary markers: Absent; content from source files is ingested without explicit delimiters instructing the agent to ignore embedded commands. (3) Capability inventory: The skill has permissions to write, edit, and execute shell commands. (4) Sanitization: Absent; content is preserved verbatim to ensure zero content loss during conversion. This is evaluated as a low-risk finding because the behavior is intrinsic to the skill's purpose as a developer-controlled formatting tool.

prompt-generator