internalcot v0.2.3

internalcot now requires a detailed visible reasoning trace while $internalcot mode is active.

Highlights:

• Requires a note before the first substantive tool call or answer.
• Records intermediate derivations, calculations, alternatives, failed approaches, uncertainty, evidence, and checks instead of concise checkpoints.
• Requires continued notes between reasoning phases and final verification before answering.
• Adds explicit non-substitution and retry/recovery instructions inspired by reliable agent-tool contracts.
• Strengthens the API observation prompt and tool schema to match the original experiment.
• Credits and links Can Bölük’s original Python proof of concept: https://pasta.can.ac/omegiligox.py

Original post: https://x.com/_can1357/status/2087228354399265125

internalcot v0.2.2

internalcot now installs a persistent working-notes skill and CLI for Codex and Claude Code.

Highlights:

• Adds guided npx internalcot@latest setup onboarding.
• Keeps $internalcot active across the current conversation with local, visible working notes.
• Supports skill-only installs through an npx fallback.
• Separates the public discovery skill from the CLI-owned runtime workflow.
• Adds Simplified Chinese documentation and preserves the API observation POC.

Credit for the original proof of concept: Can Bölük (@_can1357).