Skip to main content

simplify-gate

Show, switch on or switch off agentrc's simplify gate (the Codex YAGNI challenge that runs when a Claude Code session stops) for the current repository and all of its worktrees. `status` prints the state with the effective model and effort, `on` takes `--model`/`--effort` overrides, bare `/simplify-gate` or `help` prints the usage. `prune` frees the gate's per-user cache, dry run unless `--apply`.

ソース情報

リポジトリ
hathach/agentrc
ソースの最終更新活動
2026年10月28日 08:01
検出された SKILL.md の言語
英語
スター
2
フォーク
2

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

ファイルエクスプローラー
2 ファイル

SKILL.md を表示中

SKILL.md
ソースの指示 · 読み取り専用プレビュー
name
simplify-gate
description
Show, switch on or switch off agentrc's simplify gate (the Codex YAGNI challenge that runs when a Claude Code session stops) for the current repository and all of its worktrees. `status` prints the state with the effective model and effort, `on` takes `--model`/`--effort` overrides, bare `/simplify-gate` or `help` prints the usage. `prune` frees the gate's per-user cache, dry run unless `--apply`.
disable-model-invocation
true
# simplify-gate Run from anywhere inside the checkout and show the output verbatim: ```bash python3 <skill dir>/scripts/gate.py status python3 <skill dir>/scripts/gate.py on [--model M] [--effort E] python3 <skill dir>/scripts/gate.py off python3 <skill dir>/scripts/gate.py prune [--apply] [--days 30] [--report-days 90] python3 <skill dir>/scripts/gate.py # usage, same as `help` ``` The state is one marker file in the repository's git common dir, so every worktree sees the same switch and a new worktree inherits it. `on` without flags keeps the overrides already in the marker; `off` deletes them with the marker. The hooks run at UserPromptSubmit and Stop, so a change takes effect at the next prompt or Stop, no session restart; switched on mid-turn in a session with no saved baseline, the gate takes its baseline at that Stop and says the turn's earlier edits are not reviewed. Status shows `on`/`off`, then the model and effort with their source: `(repo)` from the marker, `(default)` from `hooks/simplify-gate/simplify_gate.py`. The hooks are registered in `~/.claude/settings.json` when the agentrc checkout's `install.py install --skill` installs the skill, or in `<dir>/.claude/settings.local.json` for `--project <dir>`; if status says on but no challenge ever runs, check that first. `prune` works on the user's gate cache (`$XDG_CACHE_HOME/agentrc/simplify-gate`), not a repository, and is a dry run until `--apply`. Sessions share one copy of each cached blob, hard-linked from the `.blobs` store; prune relinks older private copies to it after checking their content, drops the logs and unreferenced blobs of settled sessions (no queued batch, finding or error) idle over `--days`, and drops store copies no session links. A settled session whose last turn closed also loses its checkout snapshot, and with it the blobs only that snapshot held; its next prompt takes a new one. One whose turn was interrupted keeps its snapshot, the baseline its edits are reviewed against, and is listed. It never removes a session's directory or lock files, nor a blob its state still references, skips sessions a hook holds, and only reports unresolved sessions idle over `--report-days` and state it cannot read.
GitHubで見る