approval-flow

name	approval-flow
slug	approval-flow
description	Universal approval flow framework — receives structured proposals, sends to Slack channel/DM/thread, parses human approval commands (execute/partial execute/skip/reject), callbacks to execution chain. All scenario SKILLs requiring human approval should call this atomic capability, don't implement approval logic yourself.
metadata	{"openclaw":{"emoji":"✅","tier":"base","requires":{"bins":["curl"]}}}

Approval Flow — Universal Approval Flow Framework

Atomic Capability: Any scenario requiring "proposal→human approval→execution" should reuse this SKILL, don't write your own approval parsing logic.

---

Responsibility Boundaries

This SKILL is Responsible for	This SKILL is NOT Responsible for
Format proposal messages	Generate proposal content (provided by scenario SKILL)
Send to specified Slack target	Decide which channel to send to (specified by scenario SKILL)
Parse approval replies	Specific execution logic (callback to scenario SKILL)
Manage approval status	Business priority decisions

---

Input Specification

When scenario SKILLs call approval-flow, provide the following structured input:

{
  "proposal_id": "unique-id",
  "target": {
    "type": "channel | dm | thread",
    "channel_id": "CXXXXXXXXXX",
    "thread_ts": "optional-thread-ts",
    "user_id": "optional-for-dm"
  },
  "content": {
    "title": "🚨 SEO Alert — 2026-03-18",
    "body": "Formatted proposal body (Slack mrkdwn)",
    "items": [
      {"id": "①", "label": "Optimize ai-voice-generator page", "priority": "high"},
      {"id": "②", "label": "Competitor analysis report", "priority": "medium"}
    ],
    "footer": "Confirm execution? (or adjust priority)"
  },
  "config": {
    "timeout_hours": 24,
    "reminder_hours": 4,
    "auto_reject_on_timeout": true,
    "allow_partial": true
  }
}

Minimum Required: target + content.body + content.footer. Other fields have defaults.

---

Approval Command Parsing Rules

Parsing rules for human replies (matched by priority):

Human Says	Parsed As	Execution Scope
"do" / "execute" / "start" / "confirm" / "go" / "approved"	APPROVE_ALL	Execute all items
"execute ①②" / "do 1 and 3" / "only do ①"	APPROVE_PARTIAL	Execute only specified items
"skip ③" / "don't do ③"	APPROVE_EXCEPT	Execute all except specified
"don't do" / "skip" / "cancel" / "reject" / "never mind"	REJECT	Don't execute
"do ③ first" / "③ priority"	APPROVE_REORDER	Execute all, adjust order
Additional info (non-command)	INFO_UPDATE	Don't execute, update proposal context
Timeout with no reply	TIMEOUT	Handle per config

Parsing Logic

1. Remove @mentions, emoji, whitespace
2. Check if contains explicit rejection words → REJECT
3. Check if contains item references (①②③ or 1/2/3 or #1/#2) → PARTIAL/EXCEPT/REORDER
4. Check if contains affirmative words → APPROVE_ALL
5. No matches → INFO_UPDATE (treat as additional info, don't execute)

Key Principle: Better to mistakenly judge as INFO_UPDATE (don't execute) than APPROVE (incorrect execution).

---

Output Specification

After parsing approval, output structured result for scenario SKILL use:

{
  "proposal_id": "unique-id",
  "decision": "APPROVE_ALL | APPROVE_PARTIAL | APPROVE_EXCEPT | APPROVE_REORDER | REJECT | TIMEOUT | INFO_UPDATE",
  "approved_items": ["①", "②"],
  "rejected_items": ["③"],
  "reorder": ["③", "①", "②"],
  "raw_reply": "do ③ first, then ①②",
  "additional_info": "Additional info provided by human (if any)"
}

---

Usage (Scenario SKILL Perspective)

Send Proposal

After scenario SKILL prepares proposal content, send using this template:

# In your scenario SKILL:

1. Prepare proposal content (your business logic)
2. Organize data per this SKILL's input specification
3. Use message tool to send to target (channel/DM/thread)
4. Add standard footer at end of message:
   "Confirm execution? (reply 'do' to execute all / 'execute ①②' for partial / 'don't do' to skip)"
5. Record proposal status as pending

Parse Reply

After receiving human reply, process per this SKILL's parsing rules:

1. Read reply text
2. Match per "Approval Command Parsing Rules" table
3. Output structured decision
4. Based on decision type:
   - APPROVE_* → Call scenario SKILL's execution logic
   - REJECT → Record skip, notify scenario SKILL
   - INFO_UPDATE → Update context, wait for next reply
   - TIMEOUT → Handle per config

State Management

// Approval state file: <scenario-skill>/state/approval-state.json
{
  "pending": false,
  "last_proposal_id": "seo-alert-20260318",
  "last_proposal_time": "2026-03-18T08:00:00Z",
  "last_decision": "APPROVE_PARTIAL",
  "today_proposals": 1,
  "max_daily_proposals": 3
}

Anti-spam Rules:

• pending == true → Don't send new proposals (wait for previous approval)
• today_proposals >= max_daily_proposals → No more sends today
• Same content won't repeat within 24h (idempotent)

---

Standard Footer Templates

Scenario SKILLs should uniformly use at proposal end:

Simple Version (when no numbered items):

Confirm execution? (reply 'do' to execute / 'don't do' to skip)

Full Version (when numbered items exist):

Confirm execution? (reply 'do' to execute all / 'execute ①②' for partial / 'don't do' to skip)

---

Relationship with Scenario SKILLs

Scenario SKILL (proactive / seo-agent / prod-release / ...)
  ├→ Business logic (perception/decision/priority)
  ├→ approval-flow (proposal sending + approval parsing + state management)
  └→ Execution layer (their respective downstream SOPs)

This SKILL is a pure process framework, containing no business logic.

---

Atomic Capability · 2026-03-18