content-fanout

# Supervised content fan-out ## Inputs Require a canonical source URL or complete source text, a campaign ID, and an **explicit** target list. Never expand `all platforms` into every connector. Read the conversation's authoritative `<connectors>` and `<skills>` blocks to identify active publishing skills. For an unconnected target, list its connection link and leave it pending; never guess a skill slug or call an unconnected loader. ## Prepare 1. Extract verifiable claims, links, media rights, product availability, and a single canonical destination. Drop unsupported performance claims. 2. Produce a channel-specific draft and an attributed link for each target: `utm_source=<channel>&utm_medium=<medium>&utm_campaign=<campaign>`. Use `social` for social/community posts and `referral` for developer directories and syndicated technical articles; use the campaign contract's registered source mapping when one exists. Preserve any existing query parameters. Keep a separate draft for each language and audience; do not copy an identical body everywhere. 3. Present a table with target account, title, complete text, link, attachment, and proposed visibility. Ask the user to select exact targets and review the final drafts before public writes. ## Publish and verify Load each target's existing skill one at a time and follow its own official API, confirmation, and readback rules. A loaded skill can narrow the active tool set; `load_skill` remains available for switching. Send one target at a time and record the returned ID or URL. On timeout, 5xx, or ambiguous result, read the target account's recent posts before any retry. Do not issue a second write merely because the first response was lost. Use `publish_artifact` once for each **verified** public URL. Report these states separately: `draft`, `submitted`, `verified`, `failed`, and `pending`. This workflow has no durable cross-run dedup database, so it must stay supervised; a scheduled service must add persistent `source_id + channel + content_hash` idempotency before unattended use. For Reddit, Hacker News, Product Hunt, and Habr, prepare platform-native drafts for manual review. Do not automate community submissions or bypass platform publication rules. For LinkedIn, X, TikTok, Instagram, and other reviewed APIs, public automation is available only after the actual app permissions are active and the target skill proves a verified result.