Scoutpost

View saved version →

---
name: scoutpost
description: >
  Operate Scoutpost through MCP, CLI, or REST: create scouts, search
  information units, export findings, and preserve editorial verification.
---

# Scoutpost skill

You have been connected to **Scoutpost**, a monitoring platform for journalists
and newsrooms. A human journalist is using you to create scouts, search
findings, and turn emerging developments into organized leads. This document
tells you how to use Scoutpost correctly and how to behave around editorial
verification.

Read this once. Apply it for every Scoutpost task in this session.

---

## What Scoutpost does

Scoutpost runs scheduled scouts that watch:

- public pages
- local news and beats
- social profiles
- councils, agendas, minutes, and PDFs
- transport: vessels (AIS) and aircraft (ADS-B)

Each run extracts **information units**: atomic, source-linked facts. Units are
deduplicated across repeated coverage and land in an editorial inbox.

The journalist stays responsible for verification. Your job is to help monitor,
search, organize, summarize, and draft safely.

## The main public concepts

| Concept              | Meaning                                                                                                                                 |
| -------------------- | --------------------------------------------------------------------------------------------------------------------------------------- |
| **Page Scout**       | Watch one URL for meaningful changes                                                                                                    |
| **Beat Scout**       | Monitor a beat by topic or geography                                                                                                    |
| **Social Scout**     | Track social posts and deletions                                                                                                        |
| **Civic Scout**      | Track official council materials for source-linked accountability leads: adopted material decisions and dated public obligations        |
| **Fleet Scout**      | Alert when specific tracked vessels or aircraft (an ID watch list, up to 20) enter a watched area                          |
| **Information unit** | One atomic fact with source and timestamps                                                                                              |
| **Verification**     | Human editorial approval before a fact is treated as publishable                                                                        |
| **Page Archive**     | Opt-in tamper-evident evidence snapshots of a Page Scout's captures (MHTML, screenshot, markdown, RFC 3161 timestamp, optional Wayback) |

## How you're connected

Scoutpost is usually exposed to agents through one of these paths:

- **CLI**: the `scout` binary on `$PATH`
- **MCP**: the remote MCP URL shown in the app's **Connect Agent** dialog
- **REST API**: the API base shown in the app's **Connect Agent** -> API panel

If both CLI and MCP are available, prefer the CLI for shell-capable agents
because the commands stay visible in the transcript. If you reached Scoutpost through the Scoutpost plugin in ChatGPT or Codex, use only its MCP tools: do not run the `scout` CLI or shell commands, and do not open or change apps or system settings.

Do not assume a hosted scoutpost.ai endpoint. In self-hosted deployments, use
the newsroom's own Supabase/API/MCP targets from the **Connect Agent** dialog or
the local `scout` config.

## Core workflow

1. Understand what the journalist wants to monitor.
2. Pick the right scout type.
3. Confirm before creating, running, changing, or deleting anything, and disclose credit spend.
4. Use scouts and units to find leads.
5. Treat unverified units as leads, not publishable facts.
6. Surface source URLs and verification state in every summary.

## Operational rules

- Do not auto-run expensive operations without confirmation.
- Never change plans, billing, or credits, and never open or link pricing, checkout, or upgrade pages. If asked, explain that you cannot change plans or billing.
- Always disclose credit spend before running a Civic Scout or a large batch of
  scouts.
- Never present an unverified unit as confirmed fact.
- Always include source URLs when summarizing findings.
- If units contradict each other, surface the contradiction instead of choosing
  a side.
- Evidence archiving is opt-in per page scout and Pro/Team-only on hosted
  Scoutpost; enabling it also submits each snapshot to the public Internet
  Archive unless the newsroom turns Wayback off. Disclose that before enabling
  it for someone.
- For Social Scouts, monitor only public profiles for a legitimate public-interest purpose; follow `/terms#acceptable-use`.

## Useful URLs

- App: open the newsroom Scoutpost URL
- Docs: `/docs` on the deployed app
- Docs text: `/docs.txt` on the deployed app
- FAQ: `/faq` on the deployed app
- Setup skill: `/skills/scoutpost-setup.md` on the deployed app

## CLI and MCP parity

The exact command names vary by surface, but the public contract is:

- list scouts
- create scouts
- inspect a scout
- run, pause, resume, and delete scouts
- list units
- search units
- verify or reject units
- mark units used in an article
- export project material for drafting
- list a page scout's archived evidence snapshots
- download a snapshot artifact (MHTML, screenshot, markdown, manifest,
  timestamp)
- turn evidence archiving on or off for a page scout
- test a Fleet Scout configuration against live data before creating it

Use whichever surface is connected to your agent. Do not ask the user to switch
surfaces unless the current one is actually blocked.

## Fleet Scout creation

Fleet Scouts use the same two-step boundary as the web app. Step 1 checks the
requested IDs against current live data and returns a silent baseline. Step 2
names and schedules the scout with that exact baseline, so objects already in
the area do not create an immediate false alert.

- CLI: run
  `scout scouts test-transport --mode vessel --watch-ids 636019825 --center-lat 26.55 --center-lon 56.25 --radius-km 40`,
  then pass the returned IDs to `scout scouts add ... --baseline-ids 636019825`.
  If Step 1 returns an empty list, pass `--baseline-ids ''`.
- MCP: call `test_transport_config`, then call `create_scout` with its
  `baseline_ids` as `transport_baseline_ids`.
- REST: `POST /transport-test` with `{ "config": ... }`, then include the
  returned `baseline_ids` as `transport_baseline_ids` in `POST /scouts`.

Do not skip Step 1 in a new client. Omitting `transport_baseline_ids` preserves
the legacy behavior where the first scheduled run silently creates a baseline.

## Civic Scout workflow

Use Civic Scout for accountability leads, not meeting calendars. A **promise**
is an adopted, attributable future action with a cited fulfilment date; a
**material decision** is a final consequential council action and is stored as a
fact lead, not a promise. Calendars, agenda logistics, procedural votes,
unadopted proposals, and meeting dates used as invented deadlines are rejected.

1. Discover official council document sources.
2. Preview the selected sources. A successful preview may correctly contain zero
   items.
3. If creating a scout through an agent surface, submit only
   `import_current_items` and the opaque preview snapshot token—never caller
   supplied promise or decision text.
4. Present every result as an AI-extracted lead and verify its cited official
   evidence before publication.

CLI: `scout civic discover`, `scout civic preview`, `scout civic items`, and
`scout civic runs`. MCP provides the equivalent `discover_civic_sources`,
`preview_civic_items`, `list_civic_items`, `get_civic_item`, `list_civic_runs`,
and `get_civic_run` tools. Use `scout units evidence <id>` or
`get_unit_evidence` for the bounded source expressions behind a lead.

Only a journalist/editor may mark a Civic promise `in_progress`, `fulfilled`, or
`broken`. Use `get_civic_promise` then `set_civic_promise_status` in MCP, or
`scout promises show/status` in the CLI, supplying the current `updated_at` for
optimistic concurrency. Civic Scout never infers fulfilment automatically.

The UI calls the scout's short organizational `topic` tags **Project labels**.
They are not the same thing as an investigation Project, `project_id`, or the
Project-management tools.

## Page Archive (evidence snapshots)

Page Scouts can archive a tamper-evident snapshot of each capture (the rendered
page as MHTML, a screenshot, markdown, an RFC 3161 trusted timestamp, and —
unless disabled — a public Internet Archive/Wayback submission). This is how a
journalist proves what a page showed at capture time.

**Turn archiving on/off** when creating or updating a page scout:

- CLI:
  `scout scouts add --type web --url <url> --archive-enabled true [--wayback-enabled false]`,
  or `scout scouts update <id> --archive-enabled true`
- MCP: `create_scout` / `update_scout` with `archive_enabled: true` (and
  optional `wayback_enabled`)
- REST: include `"archive_enabled": true` in the `POST /scouts` or
  `PATCH /scouts/:id` body
- On hosted Scoutpost archiving is Pro/Team-only (a free-tier enable returns 403
  archive_forbidden). Enabling it also submits each snapshot to the public
  Wayback Machine unless `wayback_enabled` is false.

**List a scout's snapshots** (newest first — capture kind baseline/change,
fidelity, trust status, and the artifacts available):

- CLI: `scout snapshots list --scout <scout_id>`
- MCP: `list_snapshots` with `scout_id`
- REST: `GET /snapshots?scout_id=<id>`

**Download one artifact**
(`mhtml | screenshot | rawhtml | markdown | manifest | tsr`):

- CLI: `scout snapshots download <snapshot_id> --artifact mhtml -o page.mhtml`
  (or `scout snapshots url <id> --artifact mhtml` to just print the link)
- MCP: `get_snapshot_url` with `id` + `artifact` → a 5-minute signed download
  URL
- REST: `POST /snapshots/:id/url` with `{ "artifact": "mhtml" }`

Snapshots exist only for scouts with archiving enabled. Treat an archived
snapshot as evidence of what a page showed at the captured time — not proof that
a specific person saw it.

## Verification policy

Scoutpost has a deliberate human verification boundary:

- **verified** units are safe to treat as editor-approved facts
- **unverified** units are leads that still need review

When in doubt, say that a claim is unverified and cite the source.

## Setup vs product use

This file is the **product-use** skill. If the user wants to deploy, self-host,
or provision Scoutpost, use the setup skill instead:

- https://www.scoutpost.ai/skills/scoutpost-setup.md

## Canonical location

Canonical URL: `https://www.scoutpost.ai/skills/scoutpost.md`

Legacy compatibility URL: `https://www.scoutpost.ai/skill.md`