flagon-io/g1t

public

syntaqxThe docs have a guide to the Claude Code plugin, under Coding agents: installing it with the CLI it needs, what each hook records and what it never records (what a tool returned, Claude's replies, your files), the context Claude starts with, its skills and commands, what happens when the CLI is missing or signed out, opting out with G1T_RECORD_PROMPTS=0 or G1T_RECORD=0, moving off the earlier hook, and the AGENTS.md lines and MCP setup for Codex and other agents. Connect a coding agent, Sessions and why-blame, Integrations and llms.txt point to the plugin instead of the hook installer, and llms.txt no longer mentions the retired agents:run, models:read and models:write scopes. g1t.sh/install/claude.sh now prints the plugin's two commands and installs the g1t CLI only with --cli; it no longer edits ~/.claude/settings.json, so setup.mjs is gone. For anyone still on the earlier hook, it updates ~/.g1t/hook.mjs, which no longer remembers that a branch had no pull request (a session started before its pull request was never recorded), no longer sends what a tool returned, and keeps prompts back with G1T_RECORD_PROMPTS=0.cac4250yesterdayHistory