The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill executes a local shell script (scripts/set_title.sh) to update terminal window titles via ANSI escape sequences. The script employs standard shell utilities (echo, tr, printf) and uses double-quoting for variable expansion, effectively mitigating shell injection risks within the script's execution context.
[DATA_EXFILTRATION] (SAFE): File operations are limited to creating a local directory (~/.claude/) and writing a temporary status file to store the current title. No sensitive credentials or system configuration files are accessed, and there are no network-bound operations.
[PROMPT_INJECTION] (SAFE): The instructions focus on logical task switching and title formatting. No patterns indicating attempts to bypass model safety filters, extract system prompts, or override core agent behavior were identified.
[INDIRECT_PROMPT_INJECTION] (LOW):
Ingestion points: The terminal title is generated from the high-level summary of the user's prompt.
Boundary markers: The script is invoked via a standard bash call with the title passed as a quoted argument.
Capability inventory: Script capabilities are restricted to writing to ~/.claude/ and sending control sequences to stdout.
Sanitization: The script includes a robust sanitization step that removes control characters (tr -d '\000-\037') and limits the title length to 80 characters, which prevents common terminal hijacking techniques using malicious escape characters.

terminal-title