jq

name: jq description: jq — the command-line JSON processor and its filter language for slicing, filtering, transforming, and reshaping JSON. Use when extracting fields or nested paths from JSON on the command line, filtering/select-ing arrays of objects, reshaping or building new JSON objects, emitting CSV/TSV with @csv/@tsv, getting raw unquoted strings for shell capture with jq -r, slurping (-s) and aggregating JSON (add, group_by, averages), parsing API responses, NDJSON pipelines (-c), or injecting shell values safely with --arg/--argjson/$ENV. Triggers on mentions of jq, the jq command, the jq filter language, jq -r, .foo/.[] filters, @csv/@tsv, slurp/fromjson, or "filter/transform JSON in the terminal". This is the jq CLI and filter language, NOT a generic JSON tutorial and NOT jq language bindings (jaq, gojq, PyPI jq — defer those to their own docs).

jq - Command-Line JSON Processor

Overview

jq is a small, fast command-line tool that runs a filter program over JSON. You write a tiny program (the filter), jq reads JSON value(s) from stdin or files, runs the filter once per input value, and prints each produced value back as JSON. Think of it as sed/awk for structured JSON instead of lines of text.

Key characteristics:

• A filter = a function from one JSON value to zero, one, or many JSON values
• Composable like a shell pipeline — | pipes one filter's output into the next
• Pure & functional — no mutation; "assignment" returns a modified copy
• Streaming by default — reads whitespace-separated values, runs the filter on each
• Deterministic, no TTY — pure stdin→stdout, ideal for scripts, CI, and agents

Mental model (read this first — it prevents the #1 source of confusion)

A jq program is a filter: an input stream of JSON values → an output stream of JSON values.

• The simplest filter is . (identity) — it copies input to output, pretty-printed.
• A filter can emit 0 values (empty, a failed select), 1 value (.foo), or many (.[], range(3)). When a filter emits many values, everything downstream runs once per value (a generator / backtracking model).
• | pipes, exactly like a Unix shell: f | g runs g on each output of f.
• , forks: f, g emits all of f's values, then all of g's.
• jq never mutates the input. |= and = produce a new value with the change applied.

echo '{"a":1,"b":2}' | jq '.a'          # 1            (one value)
echo '[1,2,3]'       | jq '.[]'          # 1 2 3        (a stream of three)
echo '{"a":1,"b":2}' | jq '.a, .b'       # 1 2          (, forks the stream)
echo '[1,2,3]'       | jq '[.[] | .+10]' # [11,12,13]   (| pipes; [...] re-collects)

Disambiguation: This skill documents the jq CLI and its filter language — invocation, I/O modes, the filter syntax, and the builtin library. It is not a generic JSON tutorial, and it does not cover jq-compatible reimplementations (jaq, gojq) or library bindings (Python jq, etc.) — for those, see their own docs.

Prerequisites

CRITICAL: Before proceeding, verify jq is installed and check the version:

jq --version        # prints e.g. "jq-1.7.1"

Version note: This skill is documented against jq 1.7 / 1.7.1 (the modern baseline, the big release after a 5-year gap). Everything stable at or before jq 1.6 is "bedrock" and shown unannotated. Features added later are tagged inline as (jq 1.7+) / (jq 1.8+) only where the version is sourced — see references/version-features.md for the full feature → minimum-version map. Distros vary widely (many Linux LTS still ship 1.6 or 1.7); always confirm with jq --version. If a function errors with "is not defined", it is likely newer than your jq.

Install

jq is a single self-contained binary with no runtime dependencies.

# macOS
brew install jq

# Debian/Ubuntu
sudo apt-get install jq

# Fedora/RHEL
sudo dnf install jq

# Arch
sudo pacman -S jq

# Alpine
apk add jq

# Windows
winget install jqlang.jq      # or: choco install jq / scoop install jq

# Docker (images moved to ghcr.io in 1.7)
docker run -i ghcr.io/jqlang/jq

# Static binaries / source: https://github.com/jqlang/jq/releases

Homepage https://jqlang.org; online playground https://play.jqlang.org.

Invocation at a Glance

echo '<json>' | jq '<filter>'          # most common: pipe JSON in via stdin
jq '<filter>' file.json                # read from one or more files
jq -c '<filter>' file.json             # compact output (one value per line ⇒ NDJSON)
jq -r '<filter>'                       # raw output: strings without JSON quotes (shell capture)
jq -n '<filter>'                       # no input: filter starts from null (generate / use $ARGS,$ENV)
jq -f program.jq file.json             # load the filter from a file
jq --arg name "Ada" '<filter>'         # inject a shell value as $name (a string)

Usage: jq [options] <filter> [file...]. With no files, jq reads stdin. The filter is a program in the jq language; the most useful invocation flags:

Always single-quote the filter in the shell — jq programs use $, (), [], |, ", which the shell would otherwise mangle. Inject data with --arg/--argjson/$ENV rather than string-interpolating shell values into the program (safer; no quoting or injection bugs).

Full flag reference, I/O modes, arg-passing, exit codes, and module path: references/cli.md.

Core Workflows

All examples below are runnable and verified on jq 1.7.1.

1. Pretty-print / validate JSON

jq . reformats (pretty-prints) and exits non-zero on invalid JSON — a cheap validator.

jq . file.json                  # pretty-print
jq -c . file.json               # re-emit compact (one line)
jq -e . file.json >/dev/null    # validate: exit ≠ 0 ⇒ invalid JSON

2. Extract a field / nested path (use -r for shell capture)

jq '.user.name'    file.json    # nested object index → "Ada"  (JSON-quoted)
jq '.items[0].id'  file.json    # array index
jq -r '.user.name' file.json    # raw → Ada  (no quotes — what you want for the shell)
NAME=$(jq -r '.user.name' file.json)   # capture a string into a shell variable

Indexing a missing key yields null (not an error). Use ? to suppress type errors (.a.b?) and // "default" for a fallback (see Troubleshooting).

3. Filter a list of objects with select / pull a field from each

jq -c '.[] | select(.age > 30)' people.json   # one matching object per line
jq -r '.[] | .email'            people.json    # one email per line
jq    'map(select(.active))'    people.json    # keep the array shape, filtered

.[] iterates the array into a stream; select(f) keeps a value only when f is truthy (only false and null are falsy — 0, "", [] are truthy).

4. Reshape / build new objects

jq '[.[] | {id, name: .fullName, active: (.status == "on")}]' people.json

{id} is shorthand for {id: .id}. Object construction inside [ ... ] collects every produced object back into one array. Computed keys: {(.k): .v}.

5. Emit CSV / TSV (and read line-oriented text)

# JSON array of objects → CSV rows (‑r so quotes/escapes are literal)
jq -r '.[] | [.id, .name, .amount] | @csv' file.json
jq -r '.[] | [.id, .name] | @tsv'          file.json   # tab-separated

# Lines of text → JSON array  (raw input + slurp = the whole stdin as one string)
jq -R -s 'split("\n")' file.txt

@csv/@tsv take an array and produce one delimited row, correctly quoting/escaping. Format list (no @base32 — it doesn't exist): @text @json @base64 @base64d @uri @csv @tsv @html @sh.

6. Slurp + aggregate (sum, average, group)

printf '{"x":1}\n{"x":2}\n{"x":3}\n' | jq -s 'map(.x) | add'      # 6  (slurp NDJSON → array)
jq '[.[].amount] | add'                          orders.json      # sum a field
jq 'map(.score) | add/length'                    scores.json      # average
jq 'group_by(.dept) | map({dept: .[0].dept, n: length})' rows.json  # group + count

-s (slurp) reads all input values into one array, then runs the filter once — the standard way to aggregate across an NDJSON stream or multiple files.

7. Pass shell values safely with --arg, and reuse programs with -f

# Inject a string as $name (no shell-quoting hazards); --argjson for parsed JSON
jq --arg name "Ada Lovelace" '.[] | select(.fullName == $name)' people.json
jq --argjson min 30 '.[] | select(.age >= $min)'                people.json

# Read a secret from the environment without shell interpolation
TOKEN=… jq -n '{auth: ("Bearer " + $ENV.TOKEN)}'

# Generate from nothing with -n, then save a reusable program with -f
jq -nc '{ts: now, items: ($ARGS.positional)}' --args a b c
jq -f transform.jq data.json

--arg/--argjson also populate $ARGS.named; --args/--jsonargs fill $ARGS.positional. $ENV is an object of environment variables — prefer $ENV.TOKEN over a shell "$TOKEN".

Non-Interactive / Agent Usage

jq is excellent for scripts and agents — it never prompts and is pure stdin→stdout.

# Branch on JSON content via exit code (-e sets status from the last output)
if jq -e '.ok' resp.json >/dev/null; then echo healthy; fi

# Test-and-extract with a meaningful exit code; // empty drops nulls
jq -e '.token // empty' resp.json

# NDJSON pipeline: compact out, line tooling friendly
jq -c '.[]' big.json | while read -r line; do …; done

# Pull a second file in without shell interpolation
jq --slurpfile cfg config.json '. + $cfg[0]' data.json
jq --rawfile body message.txt   '{subject:"hi", body:$body}'

# Stream a huge document leaf-by-leaf (doesn't fit in memory)
jq -n --stream 'fromstream(1 | truncate_stream(inputs))' huge.json

Agent tips: prefer -r + --arg to feed other tools without quoting bugs; use -c for NDJSON; use -e to branch on content; use --slurpfile/--rawfile/$ENV to combine inputs and secrets without interpolation; validate cheaply with jq -e . file >/dev/null. More patterns: references/cli.md.

Quick Reference

Troubleshooting

• Strings print with quotes — that's JSON. Use -r for raw output when capturing into a shell variable: NAME=$(jq -r .name f.json).
• null vs error — indexing a missing object key yields null (not an error); indexing the wrong type is an error (Cannot index array with "foo"). Suppress with ? (.a.b?); supply a fallback with // "default".
• // is "alternative", not boolean OR — a // b yields a when a produces a value that is not null and not false, else b (it also swallows errors in a). Great for defaults; boolean OR is or.
• Empty output is usually not a crash — a select/filter matched nothing, you used empty, or a ? swallowed an error. Check the filter, not the install.
• Multiple results print one per line — wrap the filter in [ ... ] to collect them into a single array.
• "X is not defined" / "X is not a valid format" — the function/format is newer than your jq (e.g. trim/trimstr/toboolean need jq 1.8+; abs/pick/debug(msgs) need jq 1.7+). Check jq --version and references/version-features.md. (@base32, toarray, and dateadd do not exist in any jq version.)
• Big-number math is lossy — since jq 1.7 the literal decimal precision is preserved on round-trip/compare, but arithmetic still uses IEEE-754 doubles (1e17 + 10 → …020). Don't rely on jq for big-integer math.
• Shell quoting — single-quote the program; inject data with --arg/--argjson, never string concatenation. --argfile was removed in jq 1.7 — use --slurpfile/--arg.
• Date math has no dateadd — do arithmetic on epoch seconds: now + 86400 | todate. Note gmtime/mktime use a broken-down array [year, month(0-11), mday, hour, min, sec, wday, yday] (month is 0-indexed).

References

For exhaustive detail, see the bundled reference files:

• references/cli.md — every command-line flag, the I/O / slurp / raw modes, argument passing & special variables ($ARGS/$ENV/$__loc__), exit codes, modules & search path, and non-interactive/agent patterns.
• references/language.md — the filter language: paths, composition (|/,), construction, operators, conditionals & error handling, reduce/foreach, variables & destructuring, functions, assignment/path semantics, string interpolation, and comments.
• references/builtins.md — the builtin library grouped by purpose: types, arrays/objects, strings & regex, the @fmt format strings, math, dates/time, streaming, SQL-style (INDEX/IN/JOIN), and debugging/control.
• references/version-features.md — feature → minimum jq version map (bedrock ≤1.6 vs. what landed in 1.7 / 1.7.1 / 1.8), removed features, and "what version do I need?" guidance.

Resources

• Help: jq --help, and the manual at https://jqlang.org/manual/
• Homepage: https://jqlang.org
• Playground: https://play.jqlang.org
• Source / releases: https://github.com/jqlang/jq