reading-overflowed-tool-outputlisted
Install: claude install-skill sblattj/agent-chandlery
## More triggers
big-tool-result (old name)
# reading-overflowed-tool-output
`btr.py` finds and slices the files Claude Code writes when a tool result or Workflow return is too big to show inline.
## Do this
```
B=${SKILL_DIR:-~/.claude/skills/reading-overflowed-tool-output}/btr.py
python3 -I $B ls [--sid SESSION] [-n 20] # newest session: tool-results/*.txt and /tmp/claude-*/.../tasks/*.output, size + kind + first line
python3 -I $B task TASKID --keys # a Workflow return: top-level keys with types and lengths
python3 -I $B task TASKID --path stats # one JSON path (a.b[0] syntax)
python3 -I $B show FILE --grep REGEX --head 20 # any overflow file
python3 -I $B show FILE --len
```
`show` and `task` unwrap the `{"result": ...}` wrapper and the `<persisted-output>` preview automatically. Read-only.
## Why it's built this way (friction mined)
- **Two places.** MCP and Bash overflows land in `~/.claude/projects/<cwd-slug>/<session>/tool-results/*.txt`; Workflow returns land in `/tmp/claude-<uid>/<cwd-slug>/<session>/tasks/<taskId>.output` (on macOS `/tmp` resolves to `/private/tmp`; set `BTR_TMP` if your temp root differs) wrapped in `{summary, logs, result}`. `ls` shows both for the newest session.
- **The inline Workflow result is truncated**, so the stats or the claims list you need are only in the `.output` file. On one swarm run: `--keys` gave `units:list[50], stats:dict(7), claims:list[57]`, `--path stats` gave the