# Live retrieval and cost controls

## Required Monid setup

This skill requires the Monid CLI for live LinkedIn profile, bio, experience, and post discovery. It does not bundle Monid, credentials, paid access, browser automation, or Glim.

Check availability:

```bash
monid --version
```

If the command is missing, ask permission to install the public CLI, then run:

```bash
npm install -g @monid-ai/cli
monid setup --client <agent-name-if-known>
```

Omit `--client` when the agent name is unknown. Do not ask for an email solely for setup.

Monid endpoints require an API key. Ask the user to create an account at [app.monid.ai](https://app.monid.ai) and generate a key at [API keys](https://app.monid.ai/access/api-keys). Offer either to let the user run the command privately or, with explicit permission, save a key they provide:

```bash
monid keys add -k <api-key> -l main
monid keys list
```

Never print, echo, log, persist in the skill, or repeat the key. Treat software installation and credential storage as separate approval moments. If installation or authentication is declined, stop live retrieval and offer to analyze public material pasted by the user.

## Live-data workflow

Run `monid discover` for each unit need: LinkedIn profile, bio/experience, and posts. Keep the queries short. Run `monid inspect` on each candidate before executing it; the current schema is authoritative. Prefer verified endpoints, one profile URL, and the smallest useful result limit. Use `monid run`, poll the run, read its hints, and record `cost.value` when returned.

Use Glim or another public-search tool only when separately installed and authorized, and only as a fallback for representative public posts that Monid cannot retrieve. Never state or imply that Glim is included.

## Preflight

Before any Monid run:

- Inspect the endpoint description and price.
- Set a hard cost cap from the user's stated limit. If none is stated, request approval before exceeding USD $0.10 total.
- Prefer one structured profile call over several overlapping calls.
- Do not retry a charged failure without explaining the expected added cost.
- Track known spend and stop before the cap, leaving margin for uncertain metering.

## Conservative limits

Quick mode:

- At most 1 Monid profile lookup.
- At most 2 search queries for posts.
- At most 5 representative posts.
- At most 8 source links in the answer.

Expanded mode:

- At most 2 Monid profile calls when the second fills a documented gap.
- At most 4 search queries for posts.
- At most 10 representative posts.
- At most 12 source links in the answer.

Do not paginate exhaustively. Stop early when new results repeat established themes.

## Source quality

Prefer the public LinkedIn profile and original post URLs. Use company biographies or credible public interviews only to corroborate professional facts. Distinguish the person's own claims from independent reporting. Avoid people-search databases, leaked datasets, contact enrichment, private email/phone discovery, and sources that expose personal data unrelated to professional outreach.

## Graceful fallback

If one Monid endpoint fails, inspect its failure and discover one alternative endpoint without repeating a charged request blindly. If post discovery still fails, use a separately available public-search fallback or continue with bio and experience only at lower confidence. If the profile cannot be verified through Monid, ask for pasted public content. Never bypass authentication, CAPTCHAs, robots controls, or platform restrictions.
