From glance
Use when about to run commands with potentially long output (>40 lines) — builds, tests, logs, kubectl, docker, grep, find. Pipes output through glance for token-efficient summaries instead of reading all tokens.
npx claudepluginhub juxt/glance --plugin glanceThis skill is limited to using the following tools:
Check if glance is available:
Provides canonical terse bash command patterns (e.g., git log --oneline, ls -1, pytest -q) to minimize context waste when skills inspect repo state, tool output, or tests.
Enforces context-mode tools (ctx_execute, ctx_execute_file) for processing large CLI outputs like logs, tests, git diffs, builds, APIs, avoiding context bloat. Provides Bash whitelist and decision tree.
Redirects verbose output from test scripts, end-to-end tests, and long-running commands to logs/ files to avoid Bash tool truncation. Monitors with tail and docker ps.
Share bugs, ideas, or general feedback.
Check if glance is available:
which glance
If missing, use uvx --from glancecli glance as the prefix for all glance commands (e.g. cmd 2>&1 | uvx --from glancecli glance).
First time: run glance help to learn the full interface.
Pipe through glance when output is unknown or likely >40 lines:
go build, npm run build, cargo build, make)go test, pytest, jest, cargo test)kubectl logs, docker logs, journalctl)kubectl get, docker ps)grep -r, find, rg) — pipe through glance first, then drill into the captured results instead of re-running with different flagsCommands with reliably short output — git status, pwd, which, echo, ls (small dirs), git log -5.
cmd 2>&1 | glance
# or with a preset for builds/tests:
cmd 2>&1 | glance -p errors
Read the footer — it gives you the capture ID and line count.
Don't jump to full output. Use targeted queries first:
glance show <id> -a LINE 5 # context around a specific line
glance show <id> -l 50-80 # line range
glance show <id> -f 'regex' # filter stored output
Pipe once, then query the stored capture multiple times with different filters/ranges — including with different tools. Avoids re-running expensive commands. Especially useful for broad exploratory searches — run one wide rg or find, capture it, then use glance show <id> -f to explore different aspects of the results.
If you lose track of the capture ID, use glance list to find it.
Always use 2>&1 — without it, stderr (where most errors go) isn't captured by the pipe.