| name | better-tools |
| description | Analyze the local Better Tools corpus of Codex tool calls, identify tool optimization opportunities and blindspots, propose new tools or skills, and suggest focused AGENTS.md/software-engineering improvements based on observed agent behavior. |
Better Tools
Use this skill when the user wants to improve tools over time from observed Codex tool usage, asks for a tool blindspot analysis, wants suggestions for new tools/MCP servers/skills/scripts, or asks how AGENTS.md should change based on recent agent behavior.
Data Source
The companion PreToolUse hook stores a rolling corpus under $PLUGIN_DATA when Codex provides it, falling back to:
~/.codex/plugin-data/plugin-better-tools/
The raw store is a capped daily JSONL spool under events/. Do not load raw JSONL into the conversation by default. The hook updates indexes/tool-index.json automatically, and better_tools.py index can regenerate a compact deduped index. Use the index and summaries because they collapse duplicate calls by input hash and normalized command pattern.
Workflow
- Run the analyzer from the plugin root:
python3 "$PLUGIN_ROOT/scripts/better_tools.py" doctor
python3 "$PLUGIN_ROOT/scripts/better_tools.py" index --days 30
python3 "$PLUGIN_ROOT/scripts/better_tools.py" summary --days 30
If $PLUGIN_ROOT is not available in the current shell, use the installed plugin path or the development checkout:
python3 /Users/johnlindquist/dev/plugin-better-tools/plugins/plugin-better-tools/scripts/better_tools.py summary --days 30
