ワンクリックで
flow-next-ralph-init
Scaffold repo-local Ralph autonomous harness under scripts/ralph/. Use when user runs /flow-next:ralph-init.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Scaffold repo-local Ralph autonomous harness under scripts/ralph/. Use when user runs /flow-next:ralph-init.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Drive any UI surface like a real user - a web app, a Chromium-backed desktop app (Electron / WebView2, reached over CDP), or a genuinely native app (macOS AppKit/SwiftUI, or a non-CDP webview) reached via the Cua Driver / Computer Use. Detects the surface, picks the best available driver, degrades gracefully. Use to navigate sites, verify deployed UI, test web or desktop apps, capture baseline screenshots, drive a sign-in flow, scrape data, fill forms, run an e2e check, or inspect current page state. Triggers on "check the page", "verify UI", "test the site", "test this app", "drive the app", "automate this desktop app", "read docs at", "look up API", "visit URL", "browse", "screenshot", "scrape", "e2e test", "login flow", "capture baseline", "see how it looks", "inspect current", "before redesign", "Electron app", "native app".
Carmack-level implementation review of changes via the configured backend. Use when asked to review code or a diff in a flow-next repo.
Open a PR with a cognitive-aid body rendered from flow-next spec state via gh. Use whenever asked to make or open a PR in a flow-next repo.
Single-tick autonomous build-loop conductor. Advances one ready spec one stage per tick, emits PILOT_VERDICT. Use when asked to pilot a spec or backlog.
Carmack-level review of a flow-next spec or plan via the configured backend. Use when asked to review a plan or spec.
Plan a feature into a flow-next spec with tasks in .flow/. Use when asked to plan, spec out, or break down work (fn-N ids).
SOC 職業分類に基づく
| name | flow-next-ralph-init |
| description | Scaffold repo-local Ralph autonomous harness under scripts/ralph/. Use when user runs /flow-next:ralph-init. |
| user-invocable | false |
Scaffold or update repo-local Ralph harness. Opt-in only.
Compare .flow/meta.json setup_version to the plugin version; on mismatch, escalate once per plugin version. Fail-open throughout: a missing jq, .flow/meta.json, or plugin manifest silently continues.
SETUP_VER=$(jq -r '.setup_version // empty' .flow/meta.json 2>/dev/null)
PLUGIN_JSON="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/.claude-plugin/plugin.json"
PLUGIN_VER=$(jq -r '.version' "$PLUGIN_JSON" 2>/dev/null || echo "unknown")
VERSION_ACK=$(jq -r '.version_ack // empty' .flow/meta.json 2>/dev/null)
if [[ -n "$SETUP_VER" && "$PLUGIN_VER" != "unknown" && "$SETUP_VER" != "$PLUGIN_VER" ]]; then
if [[ "${FLOW_RALPH:-}" == "1" || -n "${REVIEW_RECEIPT_PATH:-}" \
|| "${FLOW_AUTONOMOUS:-}" == "1" || "${ARGUMENTS:-}" == *mode:autonomous* \
|| "$VERSION_ACK" == "$PLUGIN_VER" ]]; then
echo "Local setup v${SETUP_VER} differs from plugin v${PLUGIN_VER}. Run /flow-next:setup to refresh local scripts." >&2
else
echo "FLOW_SETUP_ASK ${SETUP_VER} ${PLUGIN_VER}"
fi
fi
If the block printed a FLOW_SETUP_ASK line, before proceeding ask the user with AskUserQuestion (local setup differs from the plugin; refresh now?), offering exactly the options Refresh now, Remind me next version, Skip this run, then continue the skill whichever is chosen:
/flow-next:setup in this session (do not run setup yourself), then continue once it finishes.PJ="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/.claude-plugin/plugin.json"
PV=$(jq -r '.version' "$PJ" 2>/dev/null)
[[ -n "$PV" && "$PV" != "null" ]] && rm -f .flow/meta.json.tmp && jq --arg v "$PV" '.version_ack = $v' .flow/meta.json > .flow/meta.json.tmp && mv .flow/meta.json.tmp .flow/meta.json
Any other output (the one-line differs notice, or nothing) is non-blocking: continue.
The plugin root resolves once via the cross-platform env-var fallback (Droid uses DROID_PLUGIN_ROOT; Claude Code documents CLAUDE_PLUGIN_ROOT as its compat alias). Subsequent blocks use $PLUGIN_ROOT:
PLUGIN_ROOT="${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}"
scripts/ralph/ in the current repo.scripts/ralph/ already exists, offer to update (preserves config.env).templates/ into scripts/ralph/ (includes ralphctl.py for pause/resume/stop/status).flowctl, flowctl.cmd, flowctl.py, flowctl_bootstrap.py, flowctl-help.txt (from $PLUGIN_ROOT/scripts/) and pick-python.sh (from $PLUGIN_ROOT/scripts/lib/) into scripts/ralph/ — flat, so the resolver lands at scripts/ralph/pick-python.sh (NOT scripts/ralph/lib/) where ralph.sh and the hook wrapper source it.scripts/ralph/ralph.sh, scripts/ralph/ralph_once.sh, scripts/ralph/flowctl, and scripts/ralph/ralphctl.py.Resolve repo root: git rev-parse --show-toplevel
Check if scripts/ralph/ exists:
Detect available review backends (skip if UPDATE_MODE=1):
if command -v rpce-cli >/dev/null 2>&1 \
|| [ -x "$HOME/RepoPrompt/repoprompt_ce_cli" ] \
|| [ -x "$HOME/Library/Application Support/RepoPrompt CE/repoprompt_ce_cli" ] \
|| command -v rp-cli >/dev/null 2>&1; then HAVE_RP=1; else HAVE_RP=0; fi
HAVE_CODEX=$(which codex >/dev/null 2>&1 && echo 1 || echo 0)
HAVE_COPILOT=$(which copilot >/dev/null 2>&1 && echo 1 || echo 0)
HAVE_CURSOR=$(which cursor-agent >/dev/null 2>&1 && echo 1 || echo 0)
Determine review backend (skip if UPDATE_MODE=1):
Multiple review backends available. Which one?
a) RepoPrompt (macOS, visual builder)
b) Codex CLI (cross-platform, GPT 5.5 High)
c) GitHub Copilot CLI (cross-platform, Claude/GPT via Copilot)
d) Cursor CLI (cross-platform, runs cursor-agent; gpt-5.5-high via Cursor subscription)
(Reply: "a", "rp", "b", "codex", "c", "copilot", "d", "cursor", or just tell me)
Wait for response. Default if empty/ambiguous: prefer rp > codex > copilot > cursor.rpcodexcopilotcursornoneCopy files using bash (MUST use cp, NOT Write tool):
If UPDATE_MODE=1 (updating):
# Backup config.env
cp scripts/ralph/config.env /tmp/ralph-config-backup.env
# Update templates (preserves runs/)
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/ralph.sh" scripts/ralph/
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/ralph_once.sh" scripts/ralph/
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/prompt_plan.md" scripts/ralph/
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/prompt_work.md" scripts/ralph/
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/prompt_completion.md" scripts/ralph/
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/watch-filter.py" scripts/ralph/
cp "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/ralphctl.py" scripts/ralph/
cp "$PLUGIN_ROOT/scripts/flowctl" "$PLUGIN_ROOT/scripts/flowctl.cmd" "$PLUGIN_ROOT/scripts/flowctl.py" "$PLUGIN_ROOT/scripts/flowctl_bootstrap.py" "$PLUGIN_ROOT/scripts/flowctl-help.txt" "$PLUGIN_ROOT/scripts/lib/pick-python.sh" scripts/ralph/
mkdir -p scripts/ralph/hooks
cp "$PLUGIN_ROOT/scripts/hooks/ralph-guard.py" "$PLUGIN_ROOT/scripts/hooks/ralph-guard" scripts/ralph/hooks/
chmod +x scripts/ralph/ralph.sh scripts/ralph/ralph_once.sh scripts/ralph/flowctl scripts/ralph/ralphctl.py scripts/ralph/hooks/ralph-guard.py scripts/ralph/hooks/ralph-guard
# Restore config.env
cp /tmp/ralph-config-backup.env scripts/ralph/config.env
If UPDATE_MODE=0 (fresh install):
mkdir -p scripts/ralph/runs scripts/ralph/hooks
cp -R "$PLUGIN_ROOT/skills/flow-next-ralph-init/templates/." scripts/ralph/
cp "$PLUGIN_ROOT/scripts/flowctl" "$PLUGIN_ROOT/scripts/flowctl.cmd" "$PLUGIN_ROOT/scripts/flowctl.py" "$PLUGIN_ROOT/scripts/flowctl_bootstrap.py" "$PLUGIN_ROOT/scripts/flowctl-help.txt" "$PLUGIN_ROOT/scripts/lib/pick-python.sh" scripts/ralph/
cp "$PLUGIN_ROOT/scripts/hooks/ralph-guard.py" "$PLUGIN_ROOT/scripts/hooks/ralph-guard" scripts/ralph/hooks/
chmod +x scripts/ralph/ralph.sh scripts/ralph/ralph_once.sh scripts/ralph/flowctl scripts/ralph/ralphctl.py scripts/ralph/hooks/ralph-guard.py scripts/ralph/hooks/ralph-guard
Note: cp -R templates/. copies all files including dotfiles (.gitignore).
Edit scripts/ralph/config.env to set the chosen review backend (skip if UPDATE_MODE=1):
PLAN_REVIEW={{PLAN_REVIEW}} with PLAN_REVIEW=<chosen>WORK_REVIEW={{WORK_REVIEW}} with WORK_REVIEW=<chosen>COMPLETION_REVIEW={{COMPLETION_REVIEW}} with COMPLETION_REVIEW=<chosen>Register project hooks (agent-driven; required for the guard to fire).
Detect host (same signals as /flow-next:setup Step 0 when available; otherwise probe the settings paths below). Then Read the target file, merge the flow-next Ralph guard entries, Edit/Write the result. Never replace the whole hooks object with only our entries. Idempotent: if an entry's command already contains scripts/ralph/hooks/ralph-guard, leave that matcher group alone (or refresh the command string to the canonical form below if it drifted).
Fingerprint for "this is a flow-next Ralph guard entry": the hook command string contains scripts/ralph/hooks/ralph-guard (wrapper and/or .py fallback).
Canonical guard command (same on every host that can run bash wrappers):
if [ -f scripts/ralph/hooks/ralph-guard ]; then bash scripts/ralph/hooks/ralph-guard; elif [ -f scripts/ralph/hooks/ralph-guard.py ]; then scripts/ralph/hooks/ralph-guard.py; fi
Timeout: 5 seconds. Type: command.
.claude/settings.jsonTarget: project file .claude/settings.json (create {"hooks":{}} skeleton if missing; preserve every non-hooks key).
Merge these four event groups under hooks (Claude schema). Matchers use regex OR so Droid interop and Claude share one entry shape:
| Event | Matcher | Notes |
|---|---|---|
PreToolUse | Bash|Execute | shell (Claude Bash, Droid Execute) |
PreToolUse | Edit|Write | file tools (Claude host names) |
PostToolUse | Bash|Execute | shell |
PostToolUse | Edit|Write | file tools (receipt-path gate parity) |
Stop | (no matcher) | stop gate |
SubagentStop | (no matcher) | subagent stop gate |
Each event's array entry is one matcher group with a single hook object {type, command, timeout} using the canonical command above.
Consent gate: Claude Code's project-hooks trust prompt is the human consent surface. Do not invent a second consent ceremony. After merge, tell the user they may need to accept/trust project hooks in the host UI for them to load this session.
.factory/hooks.jsonTarget (verified against Factory hooks-reference): project file .factory/hooks.json. Prefer that path. Fallback only if the project already stores hooks under the hooks key of .factory/settings.json and has no .factory/hooks.json — merge there instead; never invent a third path.
Host-appropriate matchers for Droid (Factory's shell tool is Execute; file tools include Create / ApplyPatch). The guard body accepts the full dual-platform sets (Bash/Execute, Edit/Write/Create/ApplyPatch).
| Event | Matcher | Notes |
|---|---|---|
PreToolUse | Bash|Execute | shell |
PreToolUse | Edit|Write|Create|ApplyPatch | Droid file tools |
PostToolUse | Bash|Execute | shell |
PostToolUse | Edit|Write|Create|ApplyPatch | file tools (receipt-path gate) |
Stop | (no matcher) | stop gate |
SubagentStop | (no matcher) | subagent stop gate |
Prefer project-relative command as above (Ralph harness is repo-local). If the host requires absolute paths, rewrite with "$FACTORY_PROJECT_DIR"/scripts/ralph/hooks/... but keep the same fingerprint substring scripts/ralph/hooks/ralph-guard.
.codex/hooks.jsonCodex has no Claude-schema plugin hooks auto-load from the marketplace plugin. Project scope is .codex/hooks.json.
Codex subset (no SubagentStop; no Edit/Write matchers — Codex only intercepts shell):
| Event | Matcher |
|---|---|
PreToolUse | Bash|Execute |
PostToolUse | Bash|Execute |
Stop | (no matcher) |
Top-level JSON must be only {"hooks":{...}} — no sibling description key (Codex rejects unknown fields and disables all hooks).
If .codex/config.toml exists, ensure exactly one hooks = true under [features] (drop deprecated codex_hooks). Same normalization intent as setup's historical Codex hooks step; do it with a careful edit, not a second copy of setup's python block unless you need it.
afterFileEdit / beforeShellExecution). Scaffold scripts/ralph/ only; print that the guard will not fire on Cursor; do not invent a Cursor-format hook file..claude/settings.json).On UPDATE_MODE=1 still re-merge hooks so a project that had scaffold but lost settings entries is repaired. Skip only when every required event already has a fingerprinted entry with the canonical command.
Print next steps (run from terminal, NOT inside the agent session):
If UPDATE_MODE=1:
Ralph updated! Your config.env was preserved.
Hooks: project settings were re-merged (idempotent). Accept the host's
project-hooks trust prompt if it appears.
Run from terminal:
- ./scripts/ralph/ralph_once.sh (one iteration, observe)
- ./scripts/ralph/ralph.sh (full loop, AFK)
- ./scripts/ralph/ralphctl.py status|pause|resume|stop (run control; not flowctl)
If UPDATE_MODE=0:
Ralph initialized!
Next steps (run from terminal, NOT inside the agent session):
- Accept project-hooks trust if the host prompts (required once)
- Edit scripts/ralph/config.env to customize settings
- ./scripts/ralph/ralph_once.sh (one iteration, observe)
- ./scripts/ralph/ralph.sh (full loop, AFK)
- ./scripts/ralph/ralphctl.py status|pause|resume|stop (run control; not flowctl)
Maintenance:
- Re-run /flow-next:ralph-init after plugin updates to refresh scripts + re-merge hooks
- Uninstall: /flow-next:uninstall removes hook entries; then manually rm -rf scripts/ralph/ if desired