The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/statusline.sh executes several shell commands based on local workspace state and files. It uses git to extract branch information and project names from the current directory. It also reads from specific files like .worktree-name, ~/.ai-memory/theme.json, and ~/.claude/session-titles/ to render the status line. While these are standard behaviors for a status-line renderer, they involve processing data from the filesystem into a shell execution context.
[COMMAND_EXECUTION]: The script uses bc for floating-point arithmetic and jq for JSON parsing of both stdin and local files. Input from Claude Code's stdin is piped directly into jq to extract session metadata.
[DATA_EXPOSURE]: The skill reads session JSONL files from ~/.claude/projects/ to calculate token usage. This is a local read operation necessary for the skill's primary function (token tracking) and does not involve external transmission of the data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the workspace (e.g., git branch names, project directory names, and content of .worktree-name) and interpolates them into the status line. While this is rendered for the user and not fed back into the LLM as a prompt, it represents an injection surface where a malicious repository could influence the agent's UI output.

status-line-live