data-tools（数据工具）

Data Tools Skill

Critical Rule

NEVER use grep, sed, or awk on JSON, YAML, TOML, XML, or CSV. When the user requests these tools on structured data, refuse and use the correct tool. Explain why text tools break (multi-line values, nesting, quoted delimiters).

Tool Selection

Rules: One format? Use its tool (jq/yq/qsv). TOML/XML? Use dasel. GitHub CLI? Use gh --jq directly, never pipe to jq. Converting? dasel -f input -w FORMAT or yq -o FORMAT.

Key Patterns

jq -- JSON (no in-place; use temp file)

jq -r '.version' package.json
jq '.users[] | select(.role == "admin") | .name' users.json
jq '.version = "2.0.0"' pkg.json > pkg.json.tmp && mv pkg.json.tmp pkg.json
jq -s '.[0] * .[1]' base.json override.json  # deep merge

yq -- YAML (in-place with -i)

yq '.services.web.image' docker-compose.yml
yq -i '.services.app.image = "node:20"' docker-compose.yml
yq -o json config.yml                        # convert to JSON

dasel -- TOML / XML / Universal

dasel -f Cargo.toml '.package.version'
dasel put -f config.toml -t string -v "2.0" '.project.version'
dasel -f input.json -w yaml                  # format conversion

qsv -- CSV / TSV

qsv headers data.csv && qsv stats data.csv --everything | qsv table
qsv search -s status "active" users.csv | qsv select name,email
qsv join user_id orders.csv id users.csv     # dataset join
qsv index big.csv && qsv sample 1000 big.csv # large file workflow

GitHub CLI -- always use --jq

gh api repos/owner/repo/releases/latest --jq '.tag_name'
gh pr list --json number,title --jq '.[] | [.number, .title] | @tsv'

Anti-Patterns

# BAD → GOOD: jq -r '.version' package.json
grep '"version"' package.json | sed 's/.*"\(.*\)".*/\1/'
# BAD → GOOD: yq -i '.services.app.image = "node:20"' compose.yml
sed -i 's/image: node:.*/image: node:20/' docker-compose.yml
# BAD → GOOD: qsv select 2 data.csv
awk -F',' '{print $2}' data.csv

References

jq manual | yq docs | dasel docs | qsv docs