메뉴
Non-blocking approval flow for Permission Slip — request, optional push wake or detached watcher, heartbeat sweep backstop, end your turn, continue when woken.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | permission-slip-approvals |
| description | Non-blocking approval flow for Permission Slip — request, optional push wake or detached watcher, heartbeat sweep backstop, end your turn, continue when woken. |
Use this skill when you need human approval for a Permission Slip action and must not block the OpenClaw session while waiting.
permission-slip request .... If the response is pending, read wait_hint, wait_command, and push_wake_configured from the JSON output.push_wake_configured is true, the Permission Slip server POSTs to your OpenClaw gateway hooks on resolution. You may skip spawning a watcher and end your turn immediately.wait_command as a detached background process. Do not poll permission-slip status in a loop inside your turn.permission-slip pending on every OpenClaw heartbeat. Act on any items in resolved; this catches push wakes that were missed.Permission Slip appr_x resolved: approved — continue the task, run permission-slip status appr_x to fetch execution_result and continue the task.If request returns executed: true or status: approved, the action already ran — skip watching.
$meta)For connectors that resolve envelope metadata (Proton Mail message-targeted actions), prefer $meta constraints over plain parameter pins on from / to — agents can spoof param values, but $meta is checked against server-fetched headers.
Proton Mail examples:
# Auto-approve read_email only for Amazon auto-confirm messages
permission-slip request --action protonmail.read_email \
--standing-constraints '{"message_id":"*","folder":"*","$meta":{"from":{"$pattern":"auto-confirm@amazon.com"}}}' \
--params '{"message_id":42,"folder":"INBOX"}'
# Recipient rule: at least one To address must match (per message)
# {"message_id":"*","$meta":{"to":{"$pattern":"*@mycorp.com"}}}
$meta.from (or legacy sender) for sender rules — not a top-level from param on read_email.$meta.bcc rarely matches received mail (IMAP omits Bcc on inbox messages); do not rely on it for inbox automation.protonmail.send_email, constrain outbound to / cc / bcc as normal params; array patterns require every recipient to match.See Proton Mail connector docs for the full action table.
# Submit an action (auto-executes when a standing approval matches)
permission-slip request --action email.send --params '{"to":"user@example.com","subject":"Hi"}'
# Heartbeat backstop — run on every OpenClaw heartbeat:
permission-slip pending
# When pending without push_wake_configured — spawn watcher in background:
permission-slip watch appr_xxxxxxxx --session-key <your session key>
# After wake — fetch the outcome:
permission-slip status appr_xxxxxxxx
# One-time setup for push wakes (private tailnet URL + hooks token):
permission-slip webhook set --url http://<host>:18789/hooks --token <token>
permission-slip webhook status --test
Run one permission-slip watch <id> per pending approval when using the watcher fallback. At personal-use scale, N small background processes is fine. With push wakes configured, heartbeat pending alone is usually enough.
pending or permission-slip status <id> recovers.status in a loop — use pending on heartbeat or the watcher pattern instead.permission-slip watch appr_x --session-key agent:main:imessage
Default notify uses openclaw system event when openclaw is on PATH. With --session-key, the default template uses --mode next-heartbeat --session-key {session_key} for a reliable targeted wake (not --mode now, which can return ok without resuming an idle session).
Automate the CI fix loop — trigger GitHub Actions (CI + audit), read failure logs, fix locally, push, and repeat until green. Use when CI is red and you want a hands-off remediation loop.
Create and push a version tag for a package (mobile, cli, or web) from the latest main branch.
Perform a comprehensive, multi-round code review on a GitHub PR. Leaves inline review comments prioritized by severity (Critical/High/Medium/Low) across up to 6 rounds, with 5-minute waits between rounds to check for fixes.
Poll a GitHub PR for new comments and PR reviews and act on them autonomously. Use when the user wants to monitor a PR for feedback and have Claude implement requested changes automatically.
Continuously pull the latest main branch and resolve merge conflicts automatically. Polls every 90 seconds and exits after 10 consecutive clean cycles.