fablelisted

# Booster Fable Consult Read the sibling skill `../booster-command/SKILL.md`, then run command `fable` through that runner. Treat the rest of the user message as the Fable question. This is a single read-only advisory consult, not the multi-agent consilium protocol. For the actual model call, pass the completed prompt through stdin to `~/.claude/scripts/fable_consult.sh`. Never assemble a raw Claude CLI call; `--tools` is variadic and can swallow a following positional prompt. A local input/argument failure is not evidence that the Fable channel is unavailable. The consult may outlive the first shell-tool yield. If `exec_command` returns a `session_id`, poll that shell session with `write_stdin` until it returns the real exit code and stdout. Do not treat `functions.wait` on a yielded JavaScript `exec` cell as shell-output capture: the cell can resume with the original pending result (`exit_code` absent and empty output) while the child PTY output remains unread. A Fable consult is complete only when the shell session itself has terminated and its stdout/stderr has been captured. After the Fable call completes, invoke `python3 ~/.claude/scripts/fable_usage.py refresh-display` and include its two spend estimate lines if it prints anything. This refreshes the current UTC month from Claude/Codex transcript stores before printing. The lines are API-equivalent / credit-rate estimates, not an actual billing ledger.