> ## Documentation Index > Fetch the complete documentation index at: https://docs.synheart.ai/llms.txt > Use this file to discover all available pages before exploring further. # Agents > What your coding agents were stopped from, asked about and let through. The summary, the rules table, team decisions, your own Claude Code usage, and the team rules **Agents**, at [synapse.synheart.ai](https://synapse.synheart.ai), shows what the guard did with your coding agents' actions: what it stopped, what it asked a person about, and what an exception let through. It's built from counts only. **No command, file path or prompt is ever sent or shown.** The guard and its rules are explained in [Writing rules](https://docs.synheart.ai/synapse/rules.md). To set it up on a machine, see [Set up as a developer](https://docs.synheart.ai/synapse/developer-setup.md). ## The top of the page * **Team** or **Mine.** **Mine** shows your own agents. **Team** shows everyone's, added up, and is for Synapse Leads and Admins. Members see **Mine** only. * **Period.** **7 days** or **30 days**. * **Refresh** reloads the figures. ## Summary | Figure | What it counts | | - | - | | **Blocked actions** | Actions stopped by a guard rule (`verdict: deny`). | | **Approval requests** | Actions the agent had to ask a person about (`verdict: ask`). | | **Exceptions** | Actions let through by an exception someone granted with `synheart guard allow`. | | **Shadow checks** | Times a trial rule (`shadow: true`) would have fired. Nothing was stopped. | ## Rules **Agent actions by rule** lists each rule that did something in the period: | Column | Meaning | | - | - | | **Rule** | The rule's `id`. | | **Outcomes** | A bar split into stopped, asked and approved, and asked and not approved. | | **Stopped** | Times the rule stopped an action. | | **Asked** | Times it asked a person. | | **Approved** | How many of those asks the person approved. | | **Exceptions** | Times an exception let the action through. | | **Shadow** | Times it would have fired as a trial rule. | Search by rule, or use the chips to show only rules that **Blocked**, **Asked** or ran in **Shadow**. When a rule was asked about at least five times and approved every time, a note under the table names it and says it *may no longer need a person*. That's a candidate to relax, so developers are interrupted less. See [When rules are too strict](https://docs.synheart.ai/synapse/rules.md#when-rules-are-too-strict). > [!TIP] > > Before turning a new rule on for everyone, add `shadow: true` and watch its **Shadow** count for a week. If it would fire on work you're happy with, adjust it before it starts stopping anyone. ## By coding agent When more than one coding agent reports, such as Claude Code and another agent, **By coding agent** shows the same counts for each. See [Connect your agent](https://docs.synheart.ai/synapse/connect-agents.md). ## When nothing shows If no machine has reported yet, the page says *No agent activity reported yet* and offers the setup commands to copy: ```bash theme={null} curl -sSL https://synheart.sh/install | bash synheart login synheart guard team join synheart guard install --write ``` Then restart Claude Code. Activity appears after the next agent session. The steps are explained in [Set up as a developer](https://docs.synheart.ai/synapse/developer-setup.md). ## Team decisions On **Team**, **What Synapse advised the team's agents** shows what Synapse told agents when they asked it for context. Team totals only. | Figure | What it counts | | - | - | | **Checked context** | Times an agent checked the team's memory before acting. | | **Asked the developer** | Times Synapse saw a conflict with a recorded decision and the agent asked the developer. | | **Blocked** | Times Synapse advised the agent not to act. | | **First-pass acceptance** | The share of agent edits accepted the first time. | | **Correction cycles** | How often an edit was rejected before one was accepted. | **First-pass acceptance** and **Correction cycles** come only from explicit verdicts: developers running `synheart feedback accept` or `reject`, or choosing accept or reject on [Agent checks](https://docs.synheart.ai/synapse/ask.md). Below them: * **Most common reasons** the agents were asked or stopped. * **Decisions cited most**, the team decisions agents leaned on. * **Intent check**, when your organization uses it: how often the agent stopped to ask or flag, and whether the developer **Accepted**, **Edited** or **Overrode** it. ## Your Claude Code usage At the bottom, everyone sees their own Claude Code usage for the last 7 days: **Sessions**, **Input tokens**, **Output tokens** and **Cache tokens** (read and written). **API-equivalent value** estimates what those tokens would cost at API prices. It isn't money spent. Only you see this section. ## Team rules **Team rules** shows the rules every developer's agent follows, and their version, such as **v4**. Open it to read them as YAML. Everyone can read them. Only Synapse Admins can change them. Each save is a new version. Developers' machines pick it up within 15 minutes. For the format and an example to start from, see [A team example](https://docs.synheart.ai/synapse/rules.md#a-team-example). ## Who can see what | Section | Who | | - | - | | **Mine**, and your own Claude Code usage | Everyone with a seat | | **Team**, including team decisions | Synapse Leads and Admins | | Read the team rules | Everyone with a seat | | Change the team rules | Synapse Admins | This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.