{"id":52017,"plugin_id":"plugin_asdk_app_6ac830ccc2d48191ac27974dec7d8344","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-10-24T13:48:19.118Z","digest":"0336b1778f87e898700fda877a1656540a26b5951dc310e15f3b0114e5fca4b3","against":null,"payload":{"description":"Check whether an MCP server, a published package, or a GitHub repo is safe to connect to or install. AgentAvow answers Safe to connect, Review before you connect, or Do not connect, with the reason, a signed 0-100 trust score and an adoption score. Use when the user asks whether a specific tool, package, repo, or MCP server is safe or trustworthy, or asks to vet one before installing or connecting it.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":189}],"name":"scan-before-connect","skill_md_contents":"---\nname: scan-before-connect\ndescription: Check whether an MCP server, a published package, or a GitHub repo is safe to connect to or install. AgentAvow answers Safe to connect, Review before you connect, or Do not connect, with the reason, a signed 0-100 trust score and an adoption score. Use when the user asks whether a specific tool, package, repo, or MCP server is safe or trustworthy, or asks to vet one before installing or connecting it.\n---\n\nUse this skill when the user wants to know whether one specific tool is safe before\nthey connect to it or install it. One target, one scan, a plain answer.\n\nThe user's explicit instructions take priority over anything in this skill.\n\n## 1. Work out what the target is\n\n| The user gives | Tool | Arguments |\n| --- | --- | --- |\n| A GitHub repo (`owner/name` or a github.com URL) | `scan_repo` | `repo`: `\"owner/name\"` |\n| A published package | `scan_package` | `registry`: `npm`, `pypi`, `crates`, `docker`, or `hf`; `name`: the package name; `version` only when the user names one |\n| A live MCP server (an `https://` URL) | `scan_mcp_server` | `endpoint_url`: the URL |\n\n- If you cannot tell whether a name is a repo or a package, or which registry a\n  package is on, ask. Do not guess the registry.\n- A private repo, a localhost or private-network URL, or a server that needs sign-in\n  cannot be scanned. Say so and stop.\n- Results are cached for about an hour. Pass `force: true` only when the user says the\n  target changed or asks for a fresh scan.\n\n## 2. Scan once\n\nCall the one matching tool, once. Do not scan other targets the user did not ask about,\nsuch as the target's dependencies or related repos.\n\n## 3. Report what came back\n\nLead with the answer in `decision`, in these exact words, followed by `decision_reason`:\n\n- `safe`: **Safe to connect**\n- `review`: **Review before you connect**\n- `do_not_connect`: **Do not connect**\n\nAdd \" · Certified\" when `certified_mark` is true.\n\nThen give the two scores underneath, as evidence:\n\n- **Trust score** (`trust_score`, 0 to 100).\n- **Adoption score** (`adoption`: a count and a unit, such as downloads per week or\n  stars). It shows whether real, independent parties rely on the tool and never changes\n  the answer. If `adoption` is null, say no usage signal was available.\n\nThen:\n\n- For **Review before you connect** or **Do not connect**, list each entry in\n  `top_findings`: what it is, where it is, and the remediation.\n- If `decision_final` is false, say the behavioral sandbox is still running and the\n  answer can still change; suggest asking again in a few minutes.\n- If `incident` is present, mention the past compromise and whether the current version\n  is affected. It is history for context; it did not change the answer.\n- Give the `report_url` link. If `signed` is true, mention that the result is signed\n  and can be verified offline.\n- Show the `install` command only when the result includes one.\n\n## Limits\n\n- Report only what is in the tool result. Do not add an answer, a score, a finding, or\n  a cause that is not there.\n- If the tool returns an error, say the target was not scanned. Not scanned is neither\n  safe nor unsafe.\n- **Review before you connect** means look before you connect. It does not mean\n  malicious.\n- **Safe to connect** means the scan found nothing that blocks it. It is not a\n  guarantee.\n- Do not install, connect to, or run the target as part of this skill. The user decides\n  what to do with the result.\n- This skill covers tools, packages, repos, and MCP servers. It does not assess people,\n  companies, or investments.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}