data-tools-skill

AI agent skill for structured data manipulation. Teaches agents when and how to use dedicated tools instead of fragile text processing (grep/sed/awk) on structured formats.

Tools Covered

Core Principle

Never use grep, sed, or awk on JSON, YAML, TOML, XML, or CSV data. These tools treat structured data as flat text and break on multi-line values, nested structures, quoted strings, and encoding differences.

Installation

Marketplace (Recommended)

Add the Netresearch marketplace once, then browse and install skills:

# Claude Code
/plugin marketplace add netresearch/claude-code-marketplace
/plugin install data-tools@netresearch-claude-code-marketplace

Without a marketplace

Since Claude Code 2.1.157 a plugin directory under your personal skills directory loads on its own, including the hooks this repo ships:

mkdir -p ~/.claude/skills
git clone https://github.com/netresearch/data-tools-skill.git \
  ~/.claude/skills/data-tools

It loads as data-tools@skills-dir on the next session. Update with git -C ~/.claude/skills/data-tools pull and start a new session; remove it by deleting the directory. This route has no claude plugin update.

npx (skills.sh)

Install with any Agent Skills-compatible agent:

npx skills add https://github.com/netresearch/data-tools-skill --skill data-tools

Limitation: npx skills installs SKILL.md-based skills only. This repo also ships hooks, which it does not install — use the marketplace or the skills directory for those.

Download Release

Download the latest release and extract to your agent's skills directory.

Git Clone

git clone https://github.com/netresearch/data-tools-skill.git

Composer (PHP Projects)

composer require netresearch/data-tools-skill

Requires netresearch/composer-agent-skill-plugin.

npm (Node Projects)

npm install --save-dev \
  @netresearch/agent-skill-coordinator \
  github:netresearch/data-tools-skill

Requires @netresearch/agent-skill-coordinator, which discovers the skill in node_modules and registers it in AGENTS.md via a postinstall hook. For pnpm, also allowlist the coordinator's postinstall:

{
  "pnpm": {
    "onlyBuiltDependencies": ["@netresearch/agent-skill-coordinator"]
  }
}

Structure

skills/data-tools/
  SKILL.md                       # Main skill file (tool selection, patterns, anti-patterns)
  references/
    jq-cookbook.md                # Comprehensive jq patterns
    yq-cookbook.md                # YAML manipulation patterns
    dasel-cookbook.md             # TOML/XML/universal selector patterns
    csv-processing.md            # qsv workflows and recipes
    mlr-cookbook.md               # Miller for JSONL, DSL, stats, joins
    enforcement-hook.md           # What the PreToolUse gate denies and lets through
hooks/hooks.json                  # Registers the PreToolUse gate for the Bash tool
scripts/
  pre_bash_structured_warn.py     # The gate (Python, standard library only)
  test_pre_bash_structured_warn.py  # Its behavioural tests

The components and the hook's data flow are described in docs/ARCHITECTURE.md.

Contributing

Contributions follow the Netresearch contributing guide. pre-commit run --all-files runs the Skill Validation linters locally, stricter than CI: markdownlint checks every Markdown file (CI: the root files), and ShellCheck runs at its default style severity on every file pre-commit identifies as shell, .envrc and Build/hooks/pre-push included (CI: error, on *.sh files only). pre-commit install --install-hooks installs them as a commit hook, but pre-commit refuses while core.hooksPath is set, which .envrc does (it points git at Build/hooks); direnv users run the linters by hand instead.

Tests

The behavioural tests of the PreToolUse hook need only Python 3.10 or later:

python3 scripts/test_pre_bash_structured_warn.py

• The script runs scripts/pre_bash_structured_warn.py as a subprocess with a hook payload per case and compares the result. It covers the deny cases (serializer rewrites, field extraction from files and from gh/glab API responses), the commands that must pass (count and presence greps, cosmetic grep -n | sed, prose in PR bodies, commit messages and quoted heredocs, output already parsed by --jq), the one-time warnings and their once-per-session deduplication, that a deny is never deduplicated, and that a session id cannot steer the state file out of the temp directory.
• Each case prints one line: OK or FEHL, the case name, the expected (erwartet) and the actual (erhalten) result. The last line is ---- Fehlschlaege: N, the number of failing cases; the script exits 1 when N is not 0.
• When it finishes, the script deletes every data-tools-hook-seen-* file in the system temp directory, so warnings already shown in a running session appear once more.

In CI, the Skill Tests workflow (.github/workflows/tests.yml) runs the script on every pull request and on pushes to main.

A pull request that changes what the hook denies, warns about or lets through adds a case to scripts/test_pre_bash_structured_warn.py that fails without the change.

Governance and policies

This repository follows the Netresearch organisation policies:

• Governance: ownership, roles, how decisions are made and disputes resolved, and continuity.
• Roadmap: planned and explicitly excluded work for the coming year.
• Handling of dependency and code analysis findings: thresholds, deadlines and the exception process for dependency (SCA) and static analysis (SAST) findings.
• Secret management: how CI and release credentials are stored, accessed and rotated.
• Access roster: who holds administrative access to this repository and the organisation.

The security assurance case for this skill (threat model, trust boundaries, countermeasures and limits) is in docs/SECURITY-ASSURANCE.md.

Checks that run on pull requests in this repository:

• Every pull request: Skill Validation (lint.yml: skill structure, manifest sync, markdownlint, yamllint, actionlint, JSON syntax, version parity, ShellCheck, ruff), Eval Validation (eval-validate.yml) and Skill Tests (tests.yml).
• Pull requests to main: security.yml with Betterleaks (secret scanning), zizmor (workflow static analysis), dependency review (fails on vulnerabilities of severity high or above), Composer Audit and Opengrep SAST (failure threshold: organisation SAST rule); Harness Verification (harness-verify.yml) and Template Drift (check-template-drift.yml).

License

This project uses split licensing:

• Code (scripts, workflows, configs): MIT
• Content (skill definitions, documentation, references): CC-BY-SA-4.0

See the individual license files for full terms.