Agent Skills: jq

jq

Filter syntax works as documented. This file covers tool selection and the idioms that prevent the common failure modes.

When not to reach for jq

• CSV/TSV → xsv; YAML → yq; analytical SQL over JSON → duckdb (read_json_auto).
• No in-place editing. Write to a temp file and move it; jq '...' f.json > f.json truncates the input before jq reads it.

Idioms that matter

jq -r '.name'                     # raw string output — no surrounding quotes
jq -c '.'                         # compact; required when piping NDJSON onward
jq '.field // "default"'          # alternative operator — survives null/missing
jq '.maybe?'                      # suppress "cannot index" instead of erroring out
jq --arg env prod '.[$env]'       # inject shell values; never interpolate into the filter string
jq -n --argjson d "$JSON" '$d.x'  # no input, build from arguments
jq -r '.[] | [.a, .b] | @csv'     # @csv / @tsv / @sh handle quoting correctly
jq 'to_entries | map(...) | from_entries'   # operate on keys
jq 'walk(if type == "object" then del(.secret) else . end)'   # recursive edit

Slurp vs. stream

-s reads the whole input into one array — convenient, but loads everything into memory. For large or NDJSON input, process record-by-record instead:

jq -c 'select(.important)' stream.ndjson    # one object per line, streaming
jq --stream 'select(length == 2)' huge.json # leaf-level streaming for a single huge document

Filter early, before any group_by/sort_by, which must materialize the whole set.

Diagnosing errors

| Message | Cause | |---|---| | parse error: Invalid numeric literal | Malformed JSON — validate with jq empty file.json | | Cannot index string with string | Field access on a non-object; check with jq 'type' | | Cannot iterate over null | Missing field; use .field // [] |

jq empty file.json is the fast validity check — silence means valid. debug prints intermediate values to stderr without disturbing the output stream.