> For the complete documentation index, see [llms.txt](https://docs.datasentinel.io/manual/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.datasentinel.io/manual/ai-connector/troubleshooting/skill-not-loaded.md). # The Skill did not load The assistant answers about your databases but without the Datasentinel Skill: no banner, no next-step menu, raw output. How to tell, how to load the Skill by hand, and how to stop it from happening a **Applies to:** every assistant that uses the Datasentinel Skill alongside the MCP server. The MCP server is reachable and signed in, the tools answer, but the conversation does not behave as described in this documentation. For an assistant that cannot reach the server at all, start with [Find the cause in mcp.log](/manual/ai-connector/troubleshooting/read-the-mcp-log.md); for a sign-in that fails, see [Sign-in problems](/manual/ai-connector/troubleshooting/sign-in-problems.md). ## What you see The Datasentinel Skill loads on demand. The assistant keeps only its name and description in context, and pulls in the full instructions when it decides that your request is about Datasentinel. When that decision goes wrong, the assistant still sees the MCP tools and calls them directly, with no instructions on how to drive them or how to present the result. The answer looks plausible but misses everything the Skill adds: | Symptom | What the Skill would have done | | ------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | | The response ends on a finding, a table, or a chart, with **no numbered menu of next steps** | Every result closes with a short menu: drill-downs drawn from the data, "prove a finding", an HTML report, a PDF report. | | **No session banner** at the top of the first answer that carries data: nothing about the platform, who you are connected as, your style, your language | The banner is displayed once, before any data, and `whoami` shows it again. | | The assistant **picks a time window on its own**, or asks you whether a time is local or UTC | It states the instance and the window in one sentence before calling the platform, and offers three windows around a single point in time. | | **Raw tool output**: JSON fragments, long unfiltered tables, identifiers without names, no key points | Key points first, then compact tables and charts, in the vocabulary of your role. | | Several analyses **chained in one response**, without stopping | One result, then a pause for your choice. | | `datasentinel help` gets a **generic answer** or a question about what you mean | The usage guide, in your language. | | The assistant **answers in English** although you wrote in another language, or in a style that does not match your role | The language of your first message, and the style your role selects (TECHNICAL for a Data admin, PLAIN for a Developer). | | The assistant **runs `curl`** or another HTTP client against the platform, or offers to | Every call goes through an MCP tool; the Skill forbids any other route. | One of these alone is a hint; two or more mean the Skill is not driving the conversation. The data the assistant returned is not wrong, since it comes from the same MCP tools, but the analysis around it is missing. ## Why it happens {% tabs %} {% tab title="The request did not match" icon="comment-dots" %} The assistant applies the Skill when the request matches its description: Datasentinel, your monitoring, an instance or a database you operate, a slow or blocked query, a replica, a report. A first message that names none of that ("is prod-db slow?", "show me the sessions") can be answered from the MCP tools alone, and the assistant skips the Skill. Once it has answered that way, the following turns often continue in the same mode. {% endtab %} {% tab title="The Skill is not discovered" icon="folder-open" %} The assistant never saw the Skill, so there was nothing to load. The usual causes are a folder in a location the assistant does not read, a folder not named `datasentinel`, an archive extracted one level too deep (`datasentinel/datasentinel/SKILL.md`), a project-level install used from another directory, or a session started before the files were in place. In the desktop apps, the uploaded Skill can also be switched off, or the capability it needs can be disabled. Each assistant's configuration guide ends with the location it reads and the command that lists what it found: see [Check that the Skill is discovered](#check-that-the-skill-is-discovered). {% endtab %} {% tab title="A long conversation" icon="hourglass-half" %} The Skill was applied earlier in the session, then the host summarized the conversation to free up context. The instructions went with the summary, and the assistant goes on with the tools alone. The banner disappears from the following answers, the menu stops showing up, and the style drifts. This is the most common cause on a session that worked at the start. {% endtab %} {% tab title="Several Skills compete" icon="layer-group" %} Another Skill with a description about PostgreSQL, SQL, or databases is installed next to this one, and the assistant picks it instead. In that case the answer may carry the other Skill's conventions. Name the Datasentinel Skill in the prompt, as shown below, and the choice is no longer the assistant's to make. {% endtab %} {% endtabs %} ## Load the Skill by hand Naming the Skill forces the load, whatever the cause. Two forms work: the assistant's own Skill invocation, or a plain sentence in the prompt. | Assistant | Invocation | | ----------------------------------------------------------------------------------------- | --------------------------------------------------------------------------- | | Claude Code, Google Antigravity CLI, Mistral Vibe CLI, Hermes Agent | `/datasentinel` followed by the request, for example `/datasentinel status` | | Visual Studio Code, Cursor | `/datasentinel` in the chat input; `/` alone lists the Skills discovered | | GitHub Copilot CLI | `Use the /datasentinel skill to ...` in the prompt | | OpenAI Codex CLI | `$datasentinel`, or `/skills` to pick it from the list | | ChatGPT desktop app | `@` in the composer, then pick `datasentinel` to attach it to the request | | Deep Agents Code | `/skill:datasentinel` | | Claude desktop app, Microsoft Copilot, Copilot Studio, Google Gemini, any other assistant | Name it in the prompt, see below | The sentence form works everywhere, including the assistants with a slash command: ``` Use the datasentinel skill to check the platform status. ``` Then verify that the Skill is active: {% stepper %} {% step %} ### Run `help` `datasentinel help` (or `/datasentinel help`) needs no sign-in and no platform call. A loaded Skill answers with its usage guide: the command list, the report types, example prompts. A generic answer means the Skill is still not applied: go to [Check that the Skill is discovered](#check-that-the-skill-is-discovered). {% endstep %} {% step %} ### Run `whoami` `datasentinel whoami` displays the session banner: platform, identity, privilege, real-time access, token, Skill version and folder, style, language. It is the proof that the Skill is reading your session. If the connection is not signed in, the same command tells you how to sign in for your assistant. See [Check your session with whoami](/manual/ai-connector/use-the-assistant/whoami-command.md). {% endstep %} {% step %} ### Ask again An answer given without the Skill is not improved after the fact. Ask for it again with the Skill named: ``` Use the datasentinel skill to redo the last analysis. ``` The response now opens with the banner if it was never shown, states the instance and the window, and ends with the next-step menu. {% endstep %} {% endstepper %} {% hint style="info" %} **Loading once is enough for the session, until the next summary.** When the host compacts a long conversation, the Skill may drop out again. The signs are the same: the menu disappears. Name the Skill once more and continue. {% endhint %} ## Check that the Skill is discovered If naming the Skill changes nothing, the assistant does not see it. List the Skills it discovered, and check the location of the `datasentinel` folder against the one your assistant reads. | Assistant | List the Skills | Location of `datasentinel/SKILL.md` | Guide | | -------------------------- | ------------------------------- | ------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------- | | Claude Code | `/skills` | `~/.claude/skills/` or `.claude/skills/` in the project | [Configure Claude Code](/manual/ai-connector/connect-a-cli-assistant/configure-claude-code.md) | | GitHub Copilot CLI | `/skills list` | `~/.copilot/skills/`, `.github/skills/`, or `.agents/skills/` | [Configure GitHub Copilot CLI](/manual/ai-connector/connect-a-cli-assistant/configure-copilot-cli.md) | | Mistral Vibe CLI | `/datasentinel` autocompletes | `~/.vibe/skills/` or `.agents/skills/` | [Configure Mistral Vibe CLI](/manual/ai-connector/connect-a-cli-assistant/configure-mistral-vibe-cli.md) | | Google Antigravity CLI | `/skills` | `~/.gemini/skills/`, `~/.gemini/antigravity-cli/skills/`, or `.agents/skills/` | [Configure Google Antigravity CLI](/manual/ai-connector/connect-a-cli-assistant/configure-antigravity-cli.md) | | OpenAI Codex CLI | `/skills` | `~/.agents/skills/` or `.agents/skills/` | [Configure OpenAI Codex CLI](/manual/ai-connector/connect-a-cli-assistant/configure-codex-cli.md) | | Claude desktop app | **Customize > Skills** | Uploaded as `datasentinel-skill.zip`; the toggle must be on | [Configure the Claude desktop app](/manual/ai-connector/connect-a-desktop-app/configure-claude-desktop-windows.md) | | ChatGPT desktop app | **Skills** panel in the sidebar | `%USERPROFILE%\.agents\skills` on Windows, `$HOME/.agents/skills` on macOS | [Configure the ChatGPT desktop app](/manual/ai-connector/connect-a-desktop-app/configure-chatgpt-desktop-app.md) | | Microsoft Copilot (Cowork) | **Customize > Skills** | Uploaded as `datasentinel-skill.zip` | [Configure the Microsoft Copilot app](/manual/ai-connector/connect-a-desktop-app/configure-microsoft-copilot-app.md) | | Microsoft Copilot Studio | Skills panel of the agent | Uploaded as `datasentinel-skill.zip` | [Configure Microsoft Copilot Studio](/manual/ai-connector/connect-a-desktop-app/configure-copilot-studio.md) | | Google Gemini | **Settings > Skills** | Uploaded as `datasentinel-skill.zip`, or `SKILL.md` alone | [Configure the Google Gemini desktop app](/manual/ai-connector/connect-a-desktop-app/configure-gemini-desktop-app.md) | | Visual Studio Code | `/skills` | `~/.copilot/skills/`, `.github/skills/`, `~/.claude/skills/`, or `~/.agents/skills/` | [Configure Visual Studio Code](/manual/ai-connector/connect-an-ide/configure-vscode.md) | | Cursor | `/` in the chat input | `~/.cursor/skills/`, `.cursor/skills/`, or `.agents/skills/` | [Configure Cursor](/manual/ai-connector/connect-an-ide/configure-cursor.md) | | Deep Agents Code | `dcode skills list` | `~/.agents/skills/` or `.agents/skills/` | [Connect a compatible assistant](/manual/ai-connector/connect-a-cli-assistant/connect-a-compatible-assistant.md) | | Hermes Agent | `hermes skills list` | `~/.hermes/skills/` | [Connect a compatible assistant](/manual/ai-connector/connect-a-cli-assistant/connect-a-compatible-assistant.md) | Three checks cover most cases: * **The folder name.** It must be `datasentinel`, directly under one of the locations above, with `SKILL.md` inside it. An archive extracted into a folder of the same name gives `datasentinel/datasentinel/SKILL.md`, which no assistant reads. * **A new session.** Skills are read when the session starts. After installing or moving the folder, close the session and open a new one; in Cursor, restart the application. * **The working directory.** A project-level install (`.claude/skills/`, `.agents/skills/`, `.github/skills/`) is seen only from that project. Start the assistant in the project, or install the Skill in your home directory. For the desktop apps, also check that the Skill is switched on and that the capability it runs through is enabled. In the Claude desktop app, **Code execution and file creation** must be on under **Settings > Capabilities**. In Copilot Studio, a Skill that failed a load-time check is skipped silently: see the troubleshooting table of [Configure Microsoft Copilot Studio](/manual/ai-connector/connect-a-desktop-app/configure-copilot-studio.md). ## Keep it loaded A few habits make the automatic load reliable: * **Open with the Skill.** Start the conversation with a command (`datasentinel status`, `datasentinel whoami`) or with a sentence that names Datasentinel. The Skill is applied from the first turn and the following questions inherit it. * **Mention your monitoring.** "What does Datasentinel say about prod-db between 2 and 3pm?" matches the description; "is prod-db slow?" may not. * **Watch for the menu.** Its absence is the earliest sign that the Skill dropped out of a long session. Name the Skill again and continue. * **Keep one copy.** When two versions of the Skill are installed in two locations, the assistant may read the older one. `datasentinel version` prints the timestamp of the copy in use, and `whoami` names its folder. ## Where to go next {% content-ref url="/pages/wO1eq0rTY2OOkwRpABfh" %} [Skill commands reference](/manual/ai-connector/use-the-assistant/skill-commands.md) {% endcontent-ref %} {% content-ref url="/pages/Z2rg4pAPxxQWv2WUzclr" %} [Check your session with whoami](/manual/ai-connector/use-the-assistant/whoami-command.md) {% endcontent-ref %} {% content-ref url="/pages/Ob8mSSQNWTmvQ8yGNZKX" %} [Find the cause in mcp.log](/manual/ai-connector/troubleshooting/read-the-mcp-log.md) {% endcontent-ref %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.datasentinel.io/manual/ai-connector/troubleshooting/skill-not-loaded.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.