Show Up in the Right LinkedIn Moments with an AI Agent — Feed-Native, Light-Touch

# Interceptor — Loadout

```yaml
loadout: interceptor
tier: apex                # Apex tier (or an active Apex trial)
mcp: https://apex.leadshark.io/mcp
```

**The job:** Feed-native radar: patrol the live stream your network is posting right now, score which moments are worth being present for, and show up with the lightest right touch — a like, a reply to the one comment that matters, or a full comment when it warrants it. Your account shouldn't be active; it should be strategically present.

## Before you run — ask the user

Only ask what you don't already know — check the conversation, your memory,
and the account itself first. Confirm anything you inferred in one line
instead of re-interviewing.

1. **What moments should I jump on — topics, ICP roles, or the accounts you care about?**
2. **How light by default — a like, a reply to a comment, or a full comment?**
3. **What's your daily pace, and who or what's off-limits?**

## Steps

1. **See** — call `feed`: Patrol your home feed — the live stream your network is posting right now.
2. **Decide** — call `get_person_activity`: Score the moment: ICP fit, intent, and whether this is someone worth being present for.
3. **Act** — call `react_to_post`: Lightest touch — a like on the post while it's live and climbing.
4. **Act** — call `engage_with_comment`: Or drop into the thread: like or reply to the one comment that matters — present, not loud.
5. **Act** — call `comment_on_post`: Escalate to a substantive comment only when the moment truly warrants it — approval-gated.

## STOP — approval gate

Present everything you found and drafted BEFORE any write. Do not send,
schedule, publish, connect, or create automations until the user approves.
Edits beat re-drafts: apply their changes, don't start over.

## Never

- Write to LinkedIn without explicit approval this run.
- Invent facts, numbers, results, or people.
- Contact anyone on the do-not-engage list, or double-touch someone
  (check `list_actions` first).
- Retry around a refusal — report what was refused and why.

## Done means

- The user approved the drafts (aim for ≤2 edit rounds).
- Approved actions are queued or scheduled — verify via `list_actions`.
- Report back: what ran, what's staged, and where to monitor it.

## Why this stays account-safe

Presence, not noise: the Interceptor patrols read-only, scores every moment before it acts, and defaults to the lightest touch — most moves are a like or a short reply, paced inside your daily caps and deduped against people you've already engaged (list_actions). Anything with words is approval-gated. It waits for the right moment instead of manufacturing activity.

Using Apex, patrol my LinkedIn home feed for live moments worth showing up in. Score each by ICP fit and intent, and for the ones that matter, take the lightest right action — a like, a reply to the one comment that counts, or a substantive comment when it truly warrants it. Stay inside my daily caps, remember who I've already touched so I never double-tap, and queue anything with words for my approval.