renatocaliari/cali-social-engage

Social Engage

Separate, opt-in skill for interaction. By default it only drafts — actual publishing requires explicit confirmation every time, and on LinkedIn it doesn't even offer automated publishing (see below).

This is intentional: reading/analysis (social-listening) and action (social-engage) are two skills with different purposes, so running one never triggers the other without the user deciding.

Configuration

config.yaml (same file as social-listening, engage section):

engage:
  auto_post_x: false   # safe default. Only true if the user explicitly asks
  tone: "direct, no marketing jargon, shares real experience"

The auto_post_x flag for automated posting exists, but the default is false and should stay false unless the user asks to change it. Even with true, always confirm the final text with the user before publishing in that specific session -- the flag turns on the capability, it doesn't replace per-post confirmation.

Flow — X

1. The user points to a post from data/ranked.json (or pastes a link).
2. Draft the reply with scripts/draft_reply.py or directly — short, in the configured tone, citing real experience from the user when it makes sense (never invent experience they don't have).
3. Show the draft and ask if it's okay to post.
4. Only call scripts/post_x_reply.py after explicit confirmation in this conversation — regardless of the auto_post_x value.

Flow — LinkedIn

No automated publishing here, full stop. The account risk on LinkedIn (see social-listening/references/linkedin-safety-limits.md) isn't worth it for a low-frequency action like commenting. The flow is:

1. Draft the comment text (same process as X).
2. Open the post link for the user.
3. They paste it manually. Slower, but keeps the account safe.

Commands

python scripts/draft_reply.py --post-id <id> --ranked data/ranked.json
python scripts/post_x_reply.py --post-id <id> --text "..."   # only after confirmation