--- # Short handover for small agent artifacts: agent skills, MCP servers, workflows (n8n, Make, ComfyUI …), # prompt packs, custom GPTs/Gems, LoRAs. Same header rules as HANDOVER.md: unknown = null, "# assumed", "# OPEN: ask owner". handover_version: 0.5 format: lite title: "" summary: "" # one sentence: what does it do for whom? kind: null # agent-skill | mcp-server | workflow | prompt-pack | ai-model | plugin | collection | other platform: [] # where it runs, e.g. [claude-code, claude.ai, cursor, n8n, comfyui, chatgpt, ollama, open-webui] built_with: [] # e.g. [claude-code, by-hand] + model if known; "# assumed" is fine stuck_at: null # one line: where it got stuck, or "none – finished, never shared" / "none – drifting: …" tags: [] stack: [] completion_percent: null remaining_effort_hours: null last_activity: null ai_generated_share: null # none | some | most | all license: null publish_check: null # BLOCK | WARN | OK offer: { type: null, price_eur: null } source: null commands: # only what you actually ran install: null # e.g. "cp -r revivable-handover ~/.claude/skills/" try: null # e.g. a sample prompt or "n8n import:workflow --input=flow.json" commands_verified: null --- # ## 1. What it does 2–4 sentences. When should an agent or person use it, and when not? ## 2. How to install and trigger it Exact steps, e.g. where the SKILL.md goes, the MCP config line, how to import the workflow, which inputs it expects. Required accounts and keys – names only, never values. ## 3. Status | Part | Status | Note | |---|---|---| | … | done / partial / broken | … | What was it tested with (model, tool version, sample input)? With date. ## 4. Where it is stuck Exact symptom, what was tried, best guess at the cause. "Finished, never shared" is fine too. ## 5. What it is allowed to do Every tool, command, network call or file access it performs or tells an agent to perform: | Area | What exactly | Needs | |---|---|---| | Shell / processes | … | … | | Network | … (TLS verified?) | … | | Files | … | … | | Instructions to agents | … (sudo? skipped permissions?) | … | Say explicitly "hidden instructions: none found" or list them. ## 6. Rights and scan Own work, AI share, third-party parts with licence. Secret scan table `Finding | Location | Assessment` and the **Publish check** verdict. ## 7. Pitfalls, contradictions, agent context Things that break (versions, platforms), where docs and code disagree, and the files that carry the original agent's knowledge (`AGENTS.md`, `CLAUDE.md`, key prompts). ## Open owner questions