Integrate Community Skill Patterns

Context

Reviewed 3002 OpenClaw community skills (VoltAgent/awesome-openclaw-skills) against joelclaw’s capabilities. Identified patterns from 4 skills worth integrating — not as direct installs but as adapted implementations wired into the Inngest event bus.

Decision

Adopt the following patterns with full attribution:

1. Daily Digest (from memory-curator by 77darius77)

• Inngest cron function at 23:55 PST daily
• Reads today’s daily log from ~/.joelclaw/workspace/memory/YYYY-MM-DD.md
• Generates structured digest: Summary, Key Events, Learnings, Connections, Open Questions, Tomorrow
• Writes to ~/Vault/Daily/digests/YYYY-MM-DD-digest.md
• Emits memory/digest.created for downstream consumers

2. iMessage Channel (from imsg by steipete)

• Install imsg CLI via brew install steipete/tap/imsg
• Create pi skill at ~/.agents/skills/imsg/SKILL.md wrapping the CLI
• Wire as communication channel in gateway alongside Telegram
• Read-only by default; send requires explicit task context

3. Agent Approval System (from local-approvals by shaiss)

• Redis-backed approval state (not JSON files)
• Auto-approve learned categories, prompt for novel actions
• Gateway integration: approval requests delivered as Telegram messages
• joelclaw approvals CLI subcommand for list/approve/deny/learn
• Emits agent/approval.requested and agent/approval.resolved events

4. Atomic Fact Pattern (from knowledge-graph by safatinaztepe)

• Apply supersede-don’t-delete pattern to memory observations
• Each observation gets a versioned fact ID
• Superseded observations keep superseded_by reference
• Aligns with existing Qdrant memory_observations collection

Attribution

All patterns sourced from openclaw/skills (MIT licensed). Each implementation credits the original skill author in code comments and this ADR.

Consequences

• Daily digest adds end-of-day synthesis to memory pipeline (fills gap between raw logs and MEMORY.md)
• iMessage adds third communication channel (Telegram, email, iMessage)
• Approval system enables “JFDI with guardrails” — autonomous action with learned permissions
• Atomic facts improve memory quality by tracking provenance and supersession