bld-runtime-tokenslisted

# bld-runtime-tokens — know the budget before spending it One command, one line back, then a recommendation. This is a status check, not a project: **never turn it into a report**. ## The command `claude-monitor` comes from `/bld-setup`'s optional extras, so it may not be here. It reads the local Claude Code transcripts in `~/.claude/projects` — no network, no key, no account access. ```bash claude-monitor --once --output json --no-emoji --plan <PLAN> ``` **Substitute `<PLAN>` with the user's actual plan**: `pro`, `max5`, `max20`, `team`, or `custom` (with `--custom-limit-tokens`). Ask them once and remember it; guessing is the bug described below, not a shortcut around it. If the command is not found, say so in one line and stop. It installs with `uv tool install claude-monitor`, but that is `/bld-setup`'s job, not this skill's. Parse the JSON with whatever the environment already has. **Do not pipe it into `ConvertFrom-Json`** unless you have confirmed the shell is PowerShell: that cmdlet does not exist in bash or zsh, so on macOS and Linux the pipe fails and takes the whole reading with it. 🚩 **`--once` is mandatory.** Without it the tool is a live-refreshing TUI that never exits — the call hangs until the tool times out. There is no interactive terminal here. ⚠️ **Don't truncate the pipe** (`| Select-Object -First N`). Breaking the pipeline early kills the process and returns exit 255 with half the JSON. Capture it whole, then read fields off the object. ## Fie