| name | social-community-engagement |
| description | Drafts comment and DM responses with classification (question, complaint, praise, troll), escalation rules, and tone variants—writes under workspace/drafts/social/replies/; human sends all outbound messages unless automation is explicitly approved in USER.md or TOOLS.md. |
| metadata | {"clawdbot":{"emoji":"💬"},"openclaw":{"emoji":"💬"}} |
social-community-engagement
Community responses with risk routing—like qf-course-researcher classifies courses as Existing / Inactive / Missing, this skill classifies threads and routes escalation.
Prerequisites
Credentials & API (qf-style)
- Draft replies: No API keys. Human sends messages unless
USER.md / TOOLS.md explicitly allow automation.
- Live send (optional): Use OpenClaw-connected channels for the platform in question; token paths for direct APIs live in
workspace/INTEGRATIONS.md when you add them.
High-level Workflow
-
Classify (classification)
- One of:
question, complaint, praise, troll, partnership, security.
- Severity: low / med / high (high = legal, threats, PII).
-
Policy check
- If high or
security → no draft; output template: “ESCALATE_TO_HUMAN” + reason.
-
Draft variants (med/low only)
- 2 replies: Short (≤280 chars) and Warm (slightly longer).
- De-escalation language for complaints; no argument.
-
Facts
- Only use facts from provided brief; if unknown: “I’ll check with the team” pattern—no invented policy.
-
DM vs public
- Note if reply should move to DM (PII, order numbers).
-
Log (reply-log.csv append row)
- Date, thread id, classification, escalated Y/N, file path (optional if CSV used).
-
Handoff
- Explicit: “Human must send; automation disabled by default.”
Outputs (required)
workspace/drafts/social/replies/<YYYY-MM-DD>-<thread-id>.md with classification + drafts or ESCALATE block.
Agent Checklist
