Authors: Mihai Criveti, Alexander Wiegand Version: 0.1.0

Strips unnecessary JSON fields from API tool responses before passing them to an LLM. Uses a whitelist approach: only fields explicitly listed in dot-notation paths are retained.

• tool_post_invoke

• Configures per-tool whitelists via the webhooks list in plugin config.
• Each webhook entry specifies a name (matching the tool name) and a fields list of dot-notation paths (e.g. ["name", "address.city"]).
• Handles two result formats: plain JSON strings and MCP content[0]["text"] dicts.
• Uses recursive prefix-based pruning to support arbitrary nesting depth.
• Runs after json_repair (priority 149 vs 145) so JSON is already valid.

config:
  debug: false
  webhooks:
    - name: "search_api"
      fields:
        - "title"
        - "url"
        - "snippet"
    - name: "weather_api"
      fields:
        - "location.name"
        - "current.temp_c"
        - "current.condition.text"

• Whitelist paths are exact dot-notation matches; no glob or wildcard support.
• Lists are always recursed into (each element is pruned), but list items have no path key of their own — within a whitelisted list, items are pruned against the list's parent path.

• Add wildcard/glob path support (e.g. results.*.title).
• Add blacklist mode as an alternative to whitelist.
• Support per-tool output size limits.