json-flat-tool（JSON flat tool 搜索）

JSON Flat Tool

⚠️ MANDATORY: Use jstool instead of read/edit/cat for JSON files!

Before using read/edit on *.json files, ask yourself: "Should I use jstool instead?"

• View: python3 ~/.agents/skills/json-flat-tool/jstool.py view <file> -s
• Edit: python3 ~/.agents/skills/json-flat-tool/jstool.py set <path> <value> <file> -f
• Search: python3 ~/.agents/skills/json-flat-tool/jstool.py find <pattern> <file>

Note on hooks: This skill previously attempted to use PreToolUse hooks to intercept Read/Edit calls on JSON files. However, Claude Code hooks cannot match on tool arguments (like file extensions). According to the Claude Code Hooks Reference, matchers only support tool name patterns (e.g., "Read", "Edit"), not argument-based filtering (e.g., "read(*.json)"). Additionally, the respond hook type does not exist; valid types are: command, http, prompt, agent. For argument-level validation, hooks must use a command script that parses tool_input JSON.

A single-script tool (jstool.py) for viewing, inspecting, and editing JSON data.

Script path: ~/.claude/skills/json-flat-tool/jstool.py

Invocation

python3 ~/.claude/skills/json-flat-tool/jstool.py <command> [args]

Commands

Omit [file] to read from stdin.

view Options

-E and -L are element-aware and never cut an element in the middle.

Edit flags

• -f — Force: apply change to file. Default is preview-only.
• Without -f: shows a color-annotated diff of the original JSON with ~~~~~ underline markers at the exact change position.

Path syntax

root              root node
count             root-level key
users[0]          array element
users[0].name     nested key
root[0].key       root-array element key

Value parsing

Values are parsed as JSON first, then fall back to plain string:

Alice          → string
42             → integer
3.14           → number
true / false   → boolean
null           → null
'{"k":"v"}'    → object
'[1,2,3]'      → array
@path/to.json  → read value from file

Output format (view)

root object
users array
users[0] object
users[0].name string Alice
users[0].age integer 30
users[1].age integer (null)    ← magenta: null with inferred type
orphan unknown (null)          ← red: null, type unknown
meta object (empty)            ← dim: empty container
config object {3 keys}         ← dim: collapsed by -d
tags array [12 items]          ← dim: collapsed by -d

Colors: cyan = path, yellow = type, green = value, magenta = inferred null, red = unknown/delete, dim = empty/collapsed.

Workflow

Explore a JSON file

# Quick structure overview
python3 ~/.claude/skills/json-flat-tool/jstool.py view data.json -s

# Filter to a nested array, element-aware pagination
python3 ~/.claude/skills/json-flat-tool/jstool.py view data.json -F "data[0].bids" -E 5 -L 3

# Infer JSON Schema Draft 7
python3 ~/.claude/skills/json-flat-tool/jstool.py schema data.json --title "My API"

# Depth-limited view: collapse containers beyond 2 key levels
python3 ~/.claude/skills/json-flat-tool/jstool.py view data.json -d 2

# Combine with filter: expand just one branch while keeping others collapsed
python3 ~/.claude/skills/json-flat-tool/jstool.py view data.json -d 1 -F users

Edit a JSON file

# Preview change (default)
python3 ~/.claude/skills/json-flat-tool/jstool.py set users[0].name Bob data.json

# Apply change
python3 ~/.claude/skills/json-flat-tool/jstool.py set users[0].name Bob data.json -f

# B-style
python3 ~/.claude/skills/json-flat-tool/jstool.py "users[0].name" = Bob data.json -f

# Insert / delete
python3 ~/.claude/skills/json-flat-tool/jstool.py before users[1] '{"name":"Eve"}' data.json -f
python3 ~/.claude/skills/json-flat-tool/jstool.py del users[2] data.json -f
python3 ~/.claude/skills/json-flat-tool/jstool.py set-null users[0].age data.json -f

Set value from file (@file)

# Write a complex object to a file, then set it at a path
python3 ~/.claude/skills/json-flat-tool/jstool.py set provider.openai @/tmp/openai.json config.json -f

Clone a subtree (copy)

# Clone an existing model as the base for a new one (preview)
python3 ~/.claude/skills/json-flat-tool/jstool.py copy \
  provider.google.models.antigravity-gemini-3-pro \
  provider.google.models.my-new-model \
  config.json

# Apply
python3 ~/.claude/skills/json-flat-tool/jstool.py copy \
  provider.google.models.antigravity-gemini-3-pro \
  provider.google.models.my-new-model \
  config.json -f

Deep-merge a patch file (merge)

# patch.json only needs to contain the fields to add/update
# preview: shows which keys are added (+) and updated (~)
python3 ~/.claude/skills/json-flat-tool/jstool.py merge provider.google.models /tmp/new-models.json config.json

# Apply
python3 ~/.claude/skills/json-flat-tool/jstool.py merge provider.google.models /tmp/new-models.json config.json -f

Find nodes by pattern (find)

# Regex search in both paths and values (default)
python3 ~/.claude/skills/json-flat-tool/jstool.py find apiKey config.json

# Search path only (-k), value only (-v)
python3 ~/.claude/skills/json-flat-tool/jstool.py find apiKey config.json -k
python3 ~/.claude/skills/json-flat-tool/jstool.py find "sk-.*" config.json -v

# Case-insensitive
python3 ~/.claude/skills/json-flat-tool/jstool.py find "APIKEY" config.json -k -i

# Glob mode (full-string wildcard, use * to match anywhere)
python3 ~/.claude/skills/json-flat-tool/jstool.py find "*api*" config.json -k -g -i

find options:

-k and -v are mutually exclusive. Without either flag, both path and value are searched.

Inline / piped JSON

echo '{"name":"Alice"}' | python3 ~/.claude/skills/json-flat-tool/jstool.py view
curl https://api.example.com/data | python3 ~/.claude/skills/json-flat-tool/jstool.py schema

HTTP API & WebSocket Analysis

This skill should be invoked automatically whenever the user pastes an API response, curl output, WSS frame, or asks to analyze any protocol-level JSON payload.

Analyze an HTTP API response

# Fetch and immediately explore structure
curl -s https://api.example.com/markets | \
  python3 ~/.claude/skills/json-flat-tool/jstool.py view -s

# Infer schema from a REST endpoint
curl -s https://api.example.com/orderbook/BTC-USD | \
  python3 ~/.claude/skills/json-flat-tool/jstool.py schema --title "Orderbook API"

# Save the raw response, then explore interactively
curl -s https://api.example.com/events > /tmp/events.json
python3 ~/.claude/skills/json-flat-tool/jstool.py view /tmp/events.json -d 2

# Search for a specific field across the response
curl -s https://api.example.com/data | \
  python3 ~/.claude/skills/json-flat-tool/jstool.py find "price" -k

Analyze a WebSocket (WSS) message capture

When a user pastes a raw WebSocket frame payload or a captured .json log:

# Paste the captured WSS JSON into a temp file, then inspect
echo '<paste-wss-frame-here>' > /tmp/wss_frame.json
python3 ~/.claude/skills/json-flat-tool/jstool.py view /tmp/wss_frame.json -s

# Compare schema across multiple frames (batch via stdin)
cat wss_frames/*.json | jq -s '.' | \
  python3 ~/.claude/skills/json-flat-tool/jstool.py schema --title "WSS Message Schema"

# Find volatile / changing fields (paginate through a large frame log)
python3 ~/.claude/skills/json-flat-tool/jstool.py view wss_log.json -F "data" -E 0 -L 5

Orderbook / market-data snapshots

# Explore top-level structure
python3 ~/.claude/skills/json-flat-tool/jstool.py view orderbook.json -d 1

# Drill into bids/asks with element-aware pagination
python3 ~/.claude/skills/json-flat-tool/jstool.py view orderbook.json -F "bids" -L 10
python3 ~/.claude/skills/json-flat-tool/jstool.py view orderbook.json -F "asks" -L 10

# Infer schema to understand field types
python3 ~/.claude/skills/json-flat-tool/jstool.py schema orderbook.json --title "CLOB Orderbook"

Tips for API / protocol analysis

Inline JSON from chat

When the user pastes raw JSON directly into the conversation, save it to a temp file:

cat > /tmp/api_payload.json << 'EOF'
<paste JSON here>
EOF
python3 ~/.claude/skills/json-flat-tool/jstool.py view /tmp/api_payload.json -s

Notes

• before / after only apply to array elements, not object keys.
• To add a new key to an object, use set.
• -f without a file path prints modified JSON to stdout.
• -E / -L require -F pointing to an array path.
• Array sampling for schema inference: up to 20 elements.
• required in schema = fields present and non-empty in all sampled elements.
• @file syntax works with set and B-style (=); merge always takes a file path directly.
• copy performs a deep clone — mutations to the copy do not affect the source.
• merge for non-dict targets replaces the value entirely (patch wins).
• find -g uses fnmatch (full-string wildcard): use *api* not api* for substring matching.
• find without -g uses Python re.search (substring regex by default).
• -d N only affects the flat view display; schema inference and edit commands are unaffected.
• -d N depth counts object key traversals only — array indices ([0], [1], …) do not increment depth.