> For the complete documentation index, see [llms.txt](https://docs.nullify.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.nullify.ai/capabilities/working-with-nullify.md).

# Working with Nullify

## Overview

Nullify runs autonomously, but the best security programs are a partnership. There are moments where Nullify needs a human decision — a judgment call it shouldn't make alone, a piece of context only your team has, or an approval before it acts. This page covers how that back-and-forth works: the **action items** Nullify raises, the **chat** you can hold with it anywhere, and the **escalations** it sends when something needs attention.

## Action Items

An **action item** is a tracked unit of work that Nullify files when it genuinely cannot proceed on its own. Rather than guessing or stalling silently, it asks — and once you answer, it picks the work back up automatically. There are a few kinds:

* **Escalated finding** — Nullify wants your call on a finding before acting: approve it, reject it, or ask for changes.
* **Clarifying question** — Nullify is missing a piece of context (a policy, an environment detail, an ownership rule). Answer once, and it re-triages every finding that was waiting on that answer.
* **Unmapped user** — Nullify knows the person involved in one system (say, a GitHub handle) but needs help mapping them to another (say, their Slack account) so it can route work to them.
* **Unassigned team** — a repository or finding has no owning team. Assign one, and every unassigned finding in that scope is routed.

<figure><picture><source srcset="https://3946391995-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FId2OtLCL6O6V3XSRYxtr%2Fuploads%2Fgit-blob-45d3d3f5f67b081ec693fbf1b133ab3d3b2142e6%2Foperator-action-items-dark.png?alt=media" media="(prefers-color-scheme: dark)"><img src="https://3946391995-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FId2OtLCL6O6V3XSRYxtr%2Fuploads%2Fgit-blob-f3ddab49a4fa648f519e7d686f8d051c94153e25%2Foperator-action-items.png?alt=media" alt="Operator action items and chat"></picture><figcaption><p>Each action-item kind beside the chat.</p></figcaption></figure>

You resolve an action item simply by supplying the judgment it asked for — from the dashboard, or by replying in the Slack thread or ticket comment where it reached you. Nullify only marks an action item complete once the underlying work has actually happened, not because it *said* it would; and it never auto-closes one just because time passed. If the underlying problem goes away on its own, Nullify resolves the item for you.

You also have a few deliberate controls of your own: reopen a finding Nullify dismissed (with a reason), re-route ownership, or just answer in chat. In the dashboard you can snooze an item for a set number of days with a reason, or dismiss it with a reason such as **Not now (ask again later)** or **Decided**. A dismissed item moves to Dismissed and can be reopened later.

## Chat with Nullify

You can talk to Nullify wherever you already work — the dashboard, [Slack](https://docs.nullify.ai/connectors/slack.md), [Microsoft Teams](https://docs.nullify.ai/connectors/microsoft-teams.md), and in pull request and ticket threads. Ask it about:

* **Findings** across code, dependencies, containers, secrets, cloud posture, and pentests.
* **Remediation campaigns** and their progress.
* **Security metrics and trends.**
* [**Threat investigations**](https://docs.nullify.ai/capabilities/threat-intelligence.md) and which repositories are affected.

In the dashboard, open chat from the **Ask Nullify** bar at the bottom of the page, or with the **Ask Nullify** button next to **Settings** in the sidebar. Each chat opens as a tab in the panel. **Recent chats** reopens an earlier one, and **Open the Operator page** takes you to [Operator](#operator), which lists the same conversations. Operator has its own chat, so the **Ask Nullify** bar is hidden there.

Nullify answers in plain language and links back to the relevant place in the dashboard. From the dashboard chat it can also **act** on your behalf — reassign a pull request's reviewers, reassign, comment on, or close a ticket, rebase or promote a stalled fix PR, allowlist a finding, or create a ticket. In Slack and Teams, chat is oriented toward questions and answering Nullify's action items; for state-changing actions it points you to the dashboard.

### Approving changes

Before Nullify makes a change from dashboard chat, such as opening a PR, filing a ticket or allowlisting a finding, it shows an approval card in the conversation: what it will do, a severity, and the **Proposed changes**. Nothing happens until you choose **Approve**; **Deny** cancels the change. The card then reads **Approved**, **Denied**, **Done** or **Failed**, and a completed change shows its **Execution result**.

Chat can make changes only for a user with the `finding:write` or `config:write` permission, and configuration changes need `config:write`. Without either permission, chat answers questions but makes no changes.

The **Approvals** switch under the **Ask Nullify** message box is on by default. Turning it off lets Nullify make changes without an approval card, in both **Ask Nullify** and Operator, until you reload the dashboard.

## Operator

**Operator** is the dashboard page for working through Nullify's questions and holding longer conversations with it. The inbox on the left has two tabs, **Action Items** and **Conversations**, and the selected item or conversation opens beside it. On a narrow screen, switch between **Inbox** and **Conversation**.

The line under the page title counts the questions waiting on you, and calls out any that are stalled, blocked, snoozed or unrouted (no team and no channel was notified).

### Action Items tab

The **Action Items** tab lists open items, highest priority first. Each row shows the item's severity, kind, team and how long it has been waiting, and labels an item that is snoozed, blocked behind another question, or resolving. Search the loaded items, or filter by **Status**, **Kind**, **Team** and **Linked finding type**. The **Kind** filter uses the names in the [action-item catalog](#action-item-catalog). Set **Status** to **Completed** or **Dismissed** to see closed items.

Selecting an item opens its conversation. Nullify's first message explains what it needs, and the item's answer controls follow it. Answer with the controls, or reply in the conversation and discuss it first.

* **Not now** snoozes the item for a number of days (3, 7, 14 or 30 unless the question offers its own durations), with a reason. A snoozed item stays open, and Nullify does not nudge or report it until the snooze ends. You can still answer it in the meantime.
* The row menu offers **Open conversation**, **Copy link**, **Dismiss** for an open item, and **Reopen** for a closed item that can be reopened.
* **Action history**, at the bottom of the inbox, lists past items with the same filters and a search by linked finding ID.

Answering, snoozing, dismissing and reopening need the `finding:write` permission. Without it, an item's conversation is read-only.

The **Work Queue** card on **Overview** lists the same questions and opens each one in Operator.

### Conversations tab

The **Conversations** tab starts with **General**, where you can ask Nullify anything about your security program, followed by your past conversations, most recent first. Conversations you start from **Ask Nullify** appear here too. Search them by title, and use **Older conversations** to load more.

**New conversation** opens **General**; a conversation joins the list once you send its first message. Each conversation's menu offers **Rename** and **Delete**, which need `finding:write`. Deleting a conversation removes its messages for everyone and cannot be undone.

## Escalations

When a finding needs developer action and isn't moving — a fix PR is open but unreviewed, a fix is blocked, or a secret needs rotating — Nullify sends an **escalation**. Each one carries the finding's facts, why it needs attention, the risk and SLA, and a link to the finding, delivered as a direct message to the owner, to a channel, or both. Reply in the thread to discuss, answer, or ask for more context.

See [Slack notifications & escalations](https://docs.nullify.ai/connectors/slack/slack-notifications.md) for how to configure where escalations land. Escalation paths and the autonomous actions Nullify is allowed to take are part of [Campaigns and Strategy](https://docs.nullify.ai/capabilities/program-management.md).

## Action-item catalog

In the dashboard, action items live in [Operator](#operator). Its **Kind** filter uses the names below.

| Kind                                       | Human input                                        |
| ------------------------------------------ | -------------------------------------------------- |
| Escalated finding (`ESCALATED_FINDING`)    | Decide how a finding should proceed.               |
| Clarifying question (`TACIT_KNOWLEDGE`)    | Capture knowledge that is not yet recorded.        |
| Unmapped user (`UNMAPPED_USER`)            | Confirm identities across connected systems.       |
| Unassigned team (`UNASSIGNED_TEAM`)        | Assign ownership.                                  |
| Context question (`CONTEXT_CLARIFICATION`) | Supply missing application or business context.    |
| Strategy proposal (`STRATEGY_PROPOSAL`)    | Review and decide on a proposed security strategy. |
| Campaign proposal (`CAMPAIGN_PROPOSAL`)    | Review and decide on proposed remediation work.    |

Dashboard chat can update the security-program strategy when that surface exposes the strategy write tool and the caller has access. Ask for confirmation of the saved result; an explanation or proposed draft is not a completed write. Available actions differ by surface and are constrained by [Manager configuration](https://docs.nullify.ai/configuration/configuration-manager.md).
