Social Stack Bootstrap

---
name: Social Stack Bootstrap
description: Stand up an automated social posting stack — connect a scheduler (Postiz, Typefully, Buffer, or similar) via MCP or API, probe every tool with real calls, and produce a working/broken/manual capability map before any workflow is built on it. Use when someone says "set up automated social posting", "wire up my scheduler", "connect Postiz/Typefully/Buffer", or "automate our social media publishing". Do NOT use for choosing which platforms to be on or defining audience and pillars — that is social-media-strategy territory; for slot timing and queue depth, use posting-cadence-designer; for producing the visual assets, use social-asset-pipeline.
---

# Social Stack Bootstrap

You are standing up the operational layer of a social marketing automation. The output is not posts — it is a verified, documented stack that downstream skills (cadence design, asset pipeline, agent loop) can run on without guessing.

## Inputs to collect first

Ask for, or infer from context: the channels in scope (X, LinkedIn, Threads, Instagram, Reddit, TikTok, etc.), the scheduler(s) available or preferred, and whether an MCP server, raw API, or both exist for each scheduler. If the user has no scheduler, recommend one by channel coverage — do not default to the most famous tool; match the channel list.

## Workflow

**1. Inventory the connection.** List every tool the scheduler's MCP exposes (or every API endpoint you will rely on). Write the list down before testing anything.

**2. Probe every tool with a real call.** Do not trust tool descriptions. Call each tool with safe, read-only arguments first. For write tools: before probing, confirm a rollback path exists (a delete/update tool that can remove what the probe creates). If no rollback path exists, do NOT write-probe a live account — record the tool as **Unverified (no rollback path)** instead. An unverified write tool is a documented limitation; junk deposited in a production queue is not. Record for each tool:

- **Working** — returns expected data. Note any required-but-undocumented parameters (e.g. a list endpoint that silently returns nothing without an explicit `status` parameter).
- **Broken/unreliable** — errors, empty responses, approval failures. Note the exact error text. Mark these "retest before relying on," not permanently dead.
- **Missing** — capabilities the workflow needs that no tool provides at all.

**3. Declare the manual boundary.** Every Missing capability becomes a documented manual step, stated bluntly: "Attaching media to a scheduled draft has no tool — this is a human step in the scheduler UI." Never design a downstream workflow that silently assumes a capability you proved absent. The manual boundary list is the most valuable artifact this skill produces.