vmware-aria 2.4.0__tar.gz → 2.4.2__tar.gz

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/.claude-plugin/plugin.json

@@ -2,7 +2,7 @@
   "$schema": "https://anthropic.com/claude-code/plugin.schema.json",
   "name": "vmware-aria",
   "displayName": "VMware Aria",
-  "version": "2.4.0",
+  "version": "2.4.2",
   "description": "VMware Aria Operations: metrics, alerts, capacity, anomaly detection — 46 MCP tools.",
   "author": {
     "name": "Wei Zhou",

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/.mcp.json

@@ -5,7 +5,7 @@
       "command": "uvx",
       "args": [
         "--from",
-        "vmware-aria==2.4.0",
+        "vmware-aria==2.4.2",
         "vmware-aria-mcp"
       ]
     }

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: vmware-aria
-Version: 2.4.0
+Version: 2.4.2
 Summary: VMware Aria Operations (vRealize Operations) monitoring, alerting, and capacity planning
 Project-URL: Homepage, https://github.com/vmware-skills/VMware-Aria
 Project-URL: Repository, https://github.com/vmware-skills/VMware-Aria
@@ -15,7 +15,7 @@ Classifier: Programming Language :: Python :: 3
 Classifier: Topic :: System :: Monitoring
 Requires-Python: >=3.10
 Requires-Dist: httpx<1.0,>=0.27
-Requires-Dist: mcp[cli]<2.0,>=1.28
+Requires-Dist: mcp[cli]<2.0,>=1.30
 Requires-Dist: python-dotenv<2.0,>=1.0
 Requires-Dist: pyyaml<7.0,>=6.0
 Requires-Dist: rich<15.0,>=13.0
@@ -26,6 +26,8 @@ Description-Content-Type: text/markdown
 <!-- mcp-name: io.github.vmware-skills/vmware-aria -->
 # VMware Aria Operations MCP Skill
 
+For a read-only check of this skill with your model and agent host, use the shared [model-fit self-test](https://github.com/vmware-skills/marketplace/blob/main/MODEL_FIT_SELF_TEST.md) in the marketplace root.
+
 > **Note**: In VCF 9.0 and later, **VMware Aria Operations** has been rebranded as **VCF Operations**. This skill works against both names — the `/suite-api/` REST endpoints are unchanged.
 
 > **Author**: Wei Zhou, VMware by Broadcom — wei-wz.zhou@broadcom.com
@@ -240,7 +242,7 @@ VMs / Hosts / Clusters / Alerts / Capacity
 - Passwords loaded from env vars or `.env` file, never from `config.yaml`
 - Every MCP tool call and every CLI command that reaches Aria audit-logged to `~/.vmware/audit.db` (via vmware-policy); CLI write operations (alert acknowledge/cancel, alert notes, alert definition management, report generate/delete, resource maintenance start/end) also append to `~/.vmware-aria/audit.log`
 - API responses sanitized (control chars stripped, 500-char limit) to prevent prompt injection
-- MCP destructive writes (alert acknowledge/cancel, alert definition and report delete, maintenance start/end) take `confirm` (default `false`): without `confirm=true` they return a `blast_radius` of what would change and change nothing; `confirm=true` is refused on a blocker (e.g. a resource already in maintenance) or an unreadable field. `confirmed` is a deprecated alias
+- MCP destructive writes (alert acknowledge/cancel, alert definition and report delete, maintenance start/end) take `confirm` (default `false`): without `confirm=true` they return a `blast_radius` of what would change and change nothing; `confirm=true` is refused on a blocker (e.g. a resource already in maintenance) or an unreadable field. The old `confirmed` alias was removed in 2.4.2; a call that passes it is refused as an unknown argument
 - TLS verification is on by default; for a private CA set `SSL_CERT_FILE` to a bundle with your CA (see setup guide). `verify_ssl: false` is for isolated self-signed labs only
 
 #### Official Broadcom References

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/README-CN.md

@@ -1,6 +1,8 @@
 <!-- mcp-name: io.github.vmware-skills/vmware-aria -->
 # VMware Aria Operations MCP Skill
 
+要用自己的模型和 Agent 平台进行只读自测，请看 marketplace 根目录的[共享模型适配自测任务](https://github.com/vmware-skills/marketplace/blob/main/MODEL_FIT_SELF_TEST.md)。
+
 > **作者**: Wei Zhou, VMware by Broadcom — wei-wz.zhou@broadcom.com
 > 本项目由 VMware 工程师维护的社区项目，非 VMware 官方产品。
 > VMware 官方开发者工具请访问 [developer.broadcom.com](https://developer.broadcom.com)。
@@ -200,7 +202,7 @@ Token 为 6 小时滑动有效期（每次调用自动延长，官方规范行
 - 密码仅从环境变量或 `.env` 文件加载，不存入 `config.yaml`
 - 每次 MCP 工具调用和每条访问 Aria 的 CLI 命令记录审计日志至 `~/.vmware/audit.db`（经 vmware-policy）；CLI 写操作（告警确认/取消、告警备注、告警定义管理、报表生成/删除、资源维护开始/结束）另写 `~/.vmware-aria/audit.log`
 - API 响应经过净化处理（去除控制字符，截断至 500 字符），防止提示注入攻击
-- MCP 的破坏性写操作（告警确认/取消、删除告警定义和报表、开始/结束维护）带 `confirm` 参数（默认 `false`）：不传 `confirm=true` 时只返回将改变什么的 `blast_radius`，不做任何修改；遇到阻断项（如资源已在维护中）或有字段读不到时，`confirm=true` 会被拒绝。`confirmed` 为已弃用的别名
+- MCP 的破坏性写操作（告警确认/取消、删除告警定义和报表、开始/结束维护）带 `confirm` 参数（默认 `false`）：不传 `confirm=true` 时只返回将改变什么的 `blast_radius`，不做任何修改；遇到阻断项（如资源已在维护中）或有字段读不到时，`confirm=true` 会被拒绝。旧的 `confirmed` 别名已在 2.4.2 移除，仍传它的调用会被当作未知参数拒绝
 - 默认开启 TLS 校验；私有 CA 请把 `SSL_CERT_FILE` 指向包含你 CA 的证书包（见 setup guide）。`verify_ssl: false` 仅用于隔离的自签名实验环境
 
 <!-- usage: remote-and-exposure -->

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/README.md

@@ -1,6 +1,8 @@
 <!-- mcp-name: io.github.vmware-skills/vmware-aria -->
 # VMware Aria Operations MCP Skill
 
+For a read-only check of this skill with your model and agent host, use the shared [model-fit self-test](https://github.com/vmware-skills/marketplace/blob/main/MODEL_FIT_SELF_TEST.md) in the marketplace root.
+
 > **Note**: In VCF 9.0 and later, **VMware Aria Operations** has been rebranded as **VCF Operations**. This skill works against both names — the `/suite-api/` REST endpoints are unchanged.
 
 > **Author**: Wei Zhou, VMware by Broadcom — wei-wz.zhou@broadcom.com
@@ -215,7 +217,7 @@ VMs / Hosts / Clusters / Alerts / Capacity
 - Passwords loaded from env vars or `.env` file, never from `config.yaml`
 - Every MCP tool call and every CLI command that reaches Aria audit-logged to `~/.vmware/audit.db` (via vmware-policy); CLI write operations (alert acknowledge/cancel, alert notes, alert definition management, report generate/delete, resource maintenance start/end) also append to `~/.vmware-aria/audit.log`
 - API responses sanitized (control chars stripped, 500-char limit) to prevent prompt injection
-- MCP destructive writes (alert acknowledge/cancel, alert definition and report delete, maintenance start/end) take `confirm` (default `false`): without `confirm=true` they return a `blast_radius` of what would change and change nothing; `confirm=true` is refused on a blocker (e.g. a resource already in maintenance) or an unreadable field. `confirmed` is a deprecated alias
+- MCP destructive writes (alert acknowledge/cancel, alert definition and report delete, maintenance start/end) take `confirm` (default `false`): without `confirm=true` they return a `blast_radius` of what would change and change nothing; `confirm=true` is refused on a blocker (e.g. a resource already in maintenance) or an unreadable field. The old `confirmed` alias was removed in 2.4.2; a call that passes it is refused as an unknown argument
 - TLS verification is on by default; for a private CA set `SSL_CERT_FILE` to a bundle with your CA (see setup guide). `verify_ssl: false` is for isolated self-signed labs only
 
 #### Official Broadcom References

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/RELEASE_NOTES.md

@@ -1,3 +1,32 @@
+## v2.4.2 — p95 in metric summaries; the `confirmed` alias is gone; error hints fit the error
+
+* **Breaking for callers still sending `confirmed`:** the deprecated alias on `acknowledge_alert`, `cancel_alert`,
+  `delete_alert_definition`, `delete_report`, `start_resource_maintenance` and `end_resource_maintenance` is removed. A call that passes it is refused
+  as an unknown argument ("Extra inputs are not permitted") and changes nothing; pass `confirm=true`. A call without
+  `confirm` previews, as before.
+* **`get_resource_metrics(summary=true)` adds `p95`** per series (nearest rank), with `p95_unavailable` saying why
+  when the window holds fewer than 20 numeric points ("widen the window or use a finer interval"). Issue #22.
+  Live on 8.18.7: 288 five-minute points, `cpu|demandmhz` p95 50.2.
+* **Error hints fit the error.** A malformed id, a 400/404/409/422 the message already explains (including the
+  9.0+-only fleet / findings / PromQL endpoints on 8.x) and a missing password env var now say "The error message
+  says what to change; connectivity is not the problem" instead of sending the agent to `doctor`, which would
+  report green. Login failures (401/403), 5xx, transport, TLS and unparseable responses keep the doctor hint.
+  A TLS connection that ends early now points to service/proxy TLS logs instead of certificate trust or host/port checks.
+* **`investigate_alert` on "Objects are not receiving data from adapter instance"**: `next_step` is now
+  `list_resources(resource_kind=all, collection_status=NO_DATA_RECEIVING)` with the adapter kind to keep, which
+  names the objects that stopped reporting, instead of "no vmware-monitor tool covers this kind".
+* Setup guide: "Behind an MCP gateway" (identity per caller, read-only kept on the backends, tool-list size,
+  timeouts, session expiry). AIops#45.
+
+## v2.4.1 — a target-limited role can use this skill
+
+* **Roles (HLD §18): the target resolver raised on every call.** It read a `name` field that this skill's target
+  config does not have, so every call by a token whose role lists targets was refused as "cannot be resolved".
+  Tokens with the built-in `admin` / `reader` roles, stdio and the CLI were not affected. Found on Log-Insight
+  (same code in five skills); the new test calls the resolver on a real loaded config.
+
+* Requires `mcp` ≥ 1.30, a security floor (GHSA-84m7-p3x7-pcfv).
+
 ## v2.4.0 — roles for network callers (HLD §18)
 
 * Declares the role metadata vmware-policy ≥ 1.24.0 needs to apply **roles** to this server over HTTPS
@@ -1756,4 +1785,4 @@ Initial release of `vmware-aria` — VMware Aria Operations AI monitoring skill.
 ### Compatibility
 - Python 3.10+
 - Aria Operations 8.x (vRealize Operations 8.x)
-- Suite API v2 (`/suite-api/api/`)
+- Suite API v2 (`/suite-api/api/`)

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/openclaw.plugin.json

@@ -11,7 +11,7 @@
       "command": "uvx",
       "args": [
         "--from",
-        "vmware-aria==2.4.0",
+        "vmware-aria==2.4.2",
         "vmware-aria-mcp"
       ]
     }

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zw008/vmware-aria",
-  "version": "2.4.0",
+  "version": "2.4.2",
   "description": "VMware Aria Operations: metrics, alerts, capacity, anomaly detection — 46 MCP tools.",
   "private": true,
   "license": "MIT",

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "vmware-aria"
-version = "2.4.0"
+version = "2.4.2"
 description = "VMware Aria Operations (vRealize Operations) monitoring, alerting, and capacity planning"
 readme = "README.md"
 license = "MIT"
@@ -23,7 +23,7 @@ dependencies = [
     "rich>=13.0,<15.0",
     "pyyaml>=6.0,<7.0",
     "python-dotenv>=1.0,<2.0",
-    "mcp[cli]>=1.28,<2.0",
+    "mcp[cli]>=1.30,<2.0",
     "vmware-policy>=1.22.0,<2.0",
 ]
 

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/server.json

@@ -7,12 +7,12 @@
     "url": "https://github.com/vmware-skills/VMware-Aria",
     "source": "github"
   },
-  "version": "2.4.0",
+  "version": "2.4.2",
   "packages": [
     {
       "registryType": "pypi",
       "identifier": "vmware-aria",
-      "version": "2.4.0",
+      "version": "2.4.2",
       "runtimeHint": "uvx",
       "packageArguments": [
         {

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/skills/vmware-aria/SKILL.md

@@ -22,6 +22,8 @@ compatibility: >
 
 # VMware Aria Operations
 
+> Self-test: [guide](https://github.com/vmware-skills/marketplace/blob/main/MODEL_FIT_SELF_TEST.md).
+
 > **Disclaimer**: This is a community-maintained open-source project and is **not affiliated with, endorsed by, or sponsored by VMware, Inc. or Broadcom Inc.** "VMware" and "Aria" are trademarks of Broadcom. Source code is publicly auditable at [github.com/vmware-skills/VMware-Aria](https://github.com/vmware-skills/VMware-Aria) under the MIT license.
 
 VMware Aria Operations (vRealize Operations 8.x, VCF Operations 9.x) AI-assisted monitoring — 46 MCP tools, by category below.
@@ -49,7 +51,7 @@ VMware Aria Operations (vRealize Operations 8.x, VCF Operations 9.x) AI-assisted
 ## Quick Install
 
 ```bash
-uv tool install vmware-aria==2.4.0
+uv tool install vmware-aria==2.4.2
 vmware-aria init      # guided setup: writes config + .env (chmod 600, password grep-safe), then verifies
 vmware-aria doctor
 ```
@@ -147,7 +149,7 @@ All MCP tools accept an optional `target`. The six destructive writes take `conf
 |----------|------|:----:|-------------|
 | Resource | `list_resources` | Read | List VMs, hosts, clusters by resource kind (or `all`); `collection_status` finds objects not receiving data |
 | | `get_resource` | Read | Get resource details with health, risk, efficiency badges |
-| | `get_resource_metrics` | Read | Time-series stats; `interval_type=DAYS` for long windows; `summary=true` gives min/max/avg/latest; `missing` explains absent |
+| | `get_resource_metrics` | Read | Time-series stats; `interval_type=DAYS` for long windows; `summary=true` gives min/max/avg/p95/latest; `missing` explains absent |
 | | `get_resource_health` | Read | Badge scores; service objects add `service.available` (a badge is not service state) |
 | | `get_top_consumers` | Read | Rank by last-hour average `value` (`latest_value` = newest point) |
 | | `list_metric_keys` | Read | Keys a resource reports with name/unit and `definition`, or a kind's defined keys — look up before querying |
@@ -306,7 +308,7 @@ Set `VMWARE_ARIA_<TARGET_NAME_UPPER>_PASSWORD` (hyphens become underscores; targ
 
 ### `invalid peer certificate: UnknownIssuer` when running uvx (corporate TLS proxy)
 
-`uvx` re-resolves from PyPI on every launch, and a TLS-intercepting proxy breaks that. Run `vmware-aria mcp` after `uv tool install vmware-aria==2.4.0` (no network on launch), or set `UV_NATIVE_TLS=true` to use the system cert store.
+`uvx` re-resolves from PyPI on every launch, and a TLS-intercepting proxy breaks that. Run `vmware-aria mcp` after `uv tool install vmware-aria==2.4.2` (no network on launch), or set `UV_NATIVE_TLS=true` to use the system cert store.
 
 ## Audit & Safety
 

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/skills/vmware-aria/references/agent-guardrails.md

@@ -40,7 +40,10 @@ the whole sequence server-side and returns:
   `resource_id`), plus `resource_name`, `resource_kind`, and a `confirmed`
   flag. Every key is always present; unresolved values are explicit `null`.
 - `next_step` — the exact vmware-monitor tool and argument to call next, or
-  `null` when there is nothing confirmed to hand off.
+  `null` when there is nothing confirmed to hand off. For an "Objects are not
+  receiving data from adapter instance" alert it is vmware-aria's own
+  `list_resources` with `collection_status=NO_DATA_RECEIVING`, which names the
+  objects that stopped reporting.
 - `warnings` — why anything above is incomplete; empty on success.
 
 A resource that cannot be resolved degrades to a warning plus explicit nulls

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/skills/vmware-aria/references/capabilities.md

@@ -8,7 +8,7 @@ Each operation is classified by autonomy level per the Enterprise Harness Engine
 |:-:|---|---|---|
 | **L1** | Read-only, raw data | Always auto-run | `list_resources`, `get_resource`, `get_resource_metrics`, `list_metric_keys`, `get_resource_properties`, `get_resource_relationships`, `list_alerts`, `get_alert`, `list_alert_definitions`, `list_alert_notes`, `list_maintenance_schedules`, `get_aria_node_resources`, `list_adapters`, `search_api_operations`, `get_api_path`, capacity / badge queries |
 | **L2** | Read + analysis / recommendation | Always auto-run | `investigate_alert` (alert → confirmed affected resource), `get_alert_recommendations` (alert → definition state → prioritized recommendations), anomaly counts, top-N consumer ranking, capacity trend forecasting, rightsizing recommendations |
-| **L3** | Single write — user must approve | Only after explicit confirmation | `acknowledge_alert` (via takeownership action), `cancel_alert`, `create_alert_definition`, `set_alert_definition_state`, `delete_alert_definition`, `generate_report`, `delete_report`, `start_resource_maintenance`, `end_resource_maintenance`, `add_alert_note` *(the only writes; all auditable. On MCP, `acknowledge_alert`, `cancel_alert`, `delete_alert_definition`, `delete_report`, `start_resource_maintenance` and `end_resource_maintenance` take `confirm` (default false): without `confirm=true` they return a `blast_radius` — what the call would change — and change nothing; `confirm=true` is refused when a blocker is found or a field the radius depends on could not be read. `confirmed` is a deprecated alias. `add_alert_note` is low risk and has no gate — a note changes neither the alert nor monitoring; the CLI still asks once)* |
+| **L3** | Single write — user must approve | Only after explicit confirmation | `acknowledge_alert` (via takeownership action), `cancel_alert`, `create_alert_definition`, `set_alert_definition_state`, `delete_alert_definition`, `generate_report`, `delete_report`, `start_resource_maintenance`, `end_resource_maintenance`, `add_alert_note` *(the only writes; all auditable. On MCP, `acknowledge_alert`, `cancel_alert`, `delete_alert_definition`, `delete_report`, `start_resource_maintenance` and `end_resource_maintenance` take `confirm` (default false): without `confirm=true` they return a `blast_radius` — what the call would change — and change nothing; `confirm=true` is refused when a blocker is found or a field the radius depends on could not be read. The old `confirmed` alias was removed in 2.4.2; a call that passes it is refused as an unknown argument. `add_alert_note` is low risk and has no gate — a note changes neither the alert nor monitoring; the CLI still asks once)* |
 | **L4** | Multi-step plan / apply workflow | *N/A currently* | — *(no multi-step orchestration; Aria is observe/analyze, not configure)* |
 | **L5** | Auto-remediation from learned pattern | Pattern library only; requires `risk:low` + `reversible:true` + `repeatable:true` | *(roadmap — candidates: auto-acknowledge known-noisy alerts, auto-cancel resolved-by-event alerts)* |
 

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/skills/vmware-aria/references/cli-reference.md

@@ -93,7 +93,7 @@ Options:
   --metrics -m TEXT    Comma-separated metric keys
                        [default: cpu|usage_average,mem|usage_average]
   --hours INT          History window in hours [default: 1]
-  --summary            Per metric n/min/max/avg/latest and change points, instead of every point
+  --summary            Per metric n/min/max/avg/p95/latest and change points, instead of every point
   --interval-type TEXT Span each point covers: MINUTES, HOURS, DAYS, WEEKS, MONTHS [default: MINUTES]
   --interval-quantity INT
                        Units of --interval-type per point, at least 1 [default: 5]
@@ -146,10 +146,13 @@ only keys that `sanitize()` leaves unchanged and that are at most 200 characters
 A missing key is not a zero. (Before this change the output was a bare object keyed by metric.)
 
 With `--summary` (MCP `summary=true`) `mode` is `summary` and `metrics` is replaced by `summary`: per key `n`, `min`,
-`max`, `avg`, `latest`, `first_timestamp_ms`, `latest_timestamp_ms`, `change_count`, `change_points` (each
+`max`, `avg`, `p95`, `latest`, `first_timestamp_ms`, `latest_timestamp_ms`, `change_count`, `change_points` (each
 `{timestamp_ms, from, to}` where the value differed from the point before — e.g. `badge|health` 100 → 25; at most 50,
 the most recent kept, `change_points_truncated` says when more were dropped) and `non_numeric_points`. `missing` is
 unchanged. A continuously varying metric such as CPU usage changes at almost every point, so read `change_count` there.
+`p95` is the nearest-rank 95th percentile of the numeric points, an observed value. It is over the points as returned,
+so with `--interval-type DAYS` it is the p95 of daily values, not of the 5-minute samples. With fewer than 20 numeric
+points it is `null` and `p95_unavailable` says why.
 
 ### `vmware-aria resource health`
 

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/skills/vmware-aria/references/setup-guide.md

@@ -16,19 +16,19 @@ Complete setup and security guide for `vmware-aria`.
 ### Via uv (recommended)
 
 ```bash
-uv tool install vmware-aria==2.4.0
+uv tool install vmware-aria==2.4.2
 ```
 
 ### Via pip
 
 ```bash
-pip install vmware-aria==2.4.0
+pip install vmware-aria==2.4.2
 ```
 
 ### From source
 
 ```bash
-git clone --branch v2.4.0 https://github.com/vmware-skills/VMware-Aria.git
+git clone --branch v2.4.2 https://github.com/vmware-skills/VMware-Aria.git
 cd VMware-Aria
 pip install -e .
 ```
@@ -194,7 +194,7 @@ Add to `~/.claude.json` (or use `claude mcp add`):
 
 > v1.5.15+ recommends `vmware-aria mcp`. Pre-1.5.15 used the legacy
 > `vmware-aria-mcp` console script (still kept for backward compatibility).
-> If using `uvx --from vmware-aria==2.4.0 vmware-aria mcp` and you hit
+> If using `uvx --from vmware-aria==2.4.2 vmware-aria mcp` and you hit
 > `invalid peer certificate: UnknownIssuer` behind a corporate TLS proxy,
 > set `UV_NATIVE_TLS=true` or use the recommended form above.
 
@@ -272,6 +272,36 @@ vmware-aria mcp --transport https --host 0.0.0.0 --port 8443 \
   `--expose read`, and a separate endpoint for operators).
 - **Several people or roles on one endpoint:** run `vmware-audit roles enable` once, then give each token a role (`vmware-audit token create NAME --role reader`, or a role in `roles.yaml` naming tools and targets; vmware-policy ≥ 1.24.0). A role limits what the token may call on this endpoint; it cannot grant more than the account has. See vmware-policy's README, "Roles for network callers".
 
+## Behind an MCP gateway
+
+A gateway (one MCP endpoint in front of several servers, for example vmware-monitor, vmware-aria and
+vmware-log-insight) works with these servers as they are: each backend runs as in
+[Serving MCP over HTTPS](#serving-mcp-over-https), and the gateway is just another client. Tool names do not
+collide across vmware-monitor, vmware-aria and vmware-log-insight, so the gateway needs no prefixes. What to
+decide:
+
+- **Who the backend sees.** If the gateway calls every backend with one token of its own, every audit row says
+  `token:<gateway>` and per-caller roles stop meaning anything. To keep per-person audit and roles, have the gateway
+  **forward each caller's own `Authorization` header** to the backend. Servers on one host with the same `OPS_HOME`
+  (default `~/.vmware`) share one token file, so one token per person works on all of them. A backend session is
+  bound to the token that opened it, so the gateway must open a backend session per caller, not share one. If the gateway does
+  its own authentication instead, give it a token per backend and accept that audit identifies the gateway, not
+  the person; forward `traceparent` or `X-Request-ID` so `vmware-audit log --correlation-id ID` still ties calls to
+  one platform request.
+- **Read-only does not depend on the gateway.** Keep the layers on each backend: a read-only platform account (the
+  real boundary), `--expose read` so write tools are never offered, and deny rules in `~/.vmware/rules.yaml`. A
+  gateway misconfiguration then cannot turn into a write.
+- **Tool list size is the main cost, not the extra hop.** The hop adds milliseconds; platform calls take hundreds
+  of milliseconds to seconds. But clients put every offered tool's schema into the model's context on every turn.
+  Even with `--expose read`, the schemas of vmware-monitor, vmware-aria and vmware-log-insight together come to
+  about 30,000 tokens or more. Offer the gateway's clients a curated subset, or group
+  tools per task, rather than passing all of them through.
+- **Timeouts and sessions.** Set the gateway's per-call timeout above the longest call you expect: a call can wait
+  up to 300 s for a free session to a busy target. A backend MCP session expires after 30 minutes idle; the gateway
+  must start a new session when the backend answers 404 for an old one, as the MCP specification requires.
+- **Scaling.** Run one backend endpoint per privilege level, not one per caller. Add sessions per target with
+  `VMWARE_MCP_POOL_SIZE` before adding processes. Point the gateway's health checks at `GET /healthz`.
+
 ## Docker Deployment
 
 ### Build and run

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/tests/eval/regression/test_a_9x_only_404_is_not_a_bad_id.py

@@ -41,7 +41,9 @@ def test_404_on_an_8x_appliance_names_the_version_and_clears_the_id(monkeypatch)
 
     msg = str(exc.value)
     assert "8.6.4" in msg and "9.0" in msg
-    assert "id you passed is not the problem" in msg
+    # Pinned to the sentence vmware-policy kept when 1.24.3 dropped "the id you
+    # passed" (these tools take no id), so the test holds on either side.
+    assert "the endpoint does not exist on this version" in msg
     assert "copy an exact UUID" not in msg, "the misleading remedy is still being sent"
 
 

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/tests/eval/regression/test_metrics_summary.py

@@ -162,3 +162,63 @@ def test_mcp_summary_option(monkeypatch):
 
     assert "error" not in result, result
     assert result["summary"]["badge|health"]["latest"] == 25.0
+
+
+# Aria#22 (from AIops#45): p95 in the summary, from the points already fetched.
+
+
+@pytest.mark.unit
+def test_summary_gives_p95_by_nearest_rank():
+    from vmware_aria.ops.metric_summary import summarize_series
+
+    # 1..100: the 95th of 100 sorted values is 95, an observed value, not an interpolation.
+    summary = summarize_series(_series([float(v) for v in reversed(range(1, 101))]))
+
+    assert summary["p95"] == 95.0
+    assert summary["p95_unavailable"] is None
+
+
+@pytest.mark.unit
+def test_p95_needs_enough_points_and_says_why_when_it_has_too_few():
+    from vmware_aria.ops import metric_summary
+
+    enough = metric_summary.summarize_series(_series([float(v) for v in range(metric_summary.MIN_POINTS_FOR_P95)]))
+    too_few = metric_summary.summarize_series(_series([float(v) for v in range(metric_summary.MIN_POINTS_FOR_P95 - 1)]))
+
+    assert enough["p95"] is not None
+    assert too_few["p95"] is None
+    assert str(metric_summary.MIN_POINTS_FOR_P95) in too_few["p95_unavailable"]
+
+
+@pytest.mark.unit
+def test_p95_of_no_points_is_unknown_not_zero():
+    from vmware_aria.ops.metric_summary import summarize_series
+
+    summary = summarize_series([{"timestamp_ms": T0, "value": None}])
+
+    assert summary["p95"] is None
+    assert summary["p95_unavailable"]
+
+
+@pytest.mark.unit
+def test_p95_ignores_points_that_are_not_numbers():
+    from vmware_aria.ops import metric_summary
+
+    # 20 numbers 1..20 and 30 non-numeric points: p95 is rank ceil(0.95 * 20) = 19 of the
+    # numbers alone. Counting the 30 would move it.
+    points = _series([float(v) for v in range(1, 21)]) + [{"timestamp_ms": T0 + 99 * STEP, "value": None}] * 30
+    summary = metric_summary.summarize_series(points)
+
+    assert summary["n"] == 20
+    assert summary["p95"] == 19.0
+
+
+@pytest.mark.unit
+def test_get_resource_metrics_summary_carries_p95():
+    from vmware_aria.ops.resources import get_resource_metrics
+
+    client = MagicMock()
+    client.post.return_value = _stats_body({"cpu|usage_average": [float(v) for v in range(1, 101)]})
+    result = get_resource_metrics(client, RID, ["cpu|usage_average"], summary=True)
+
+    assert result["summary"]["cpu|usage_average"]["p95"] == 95.0

vmware_aria-2.4.2/tests/test_error_hints.py

@@ -0,0 +1,52 @@
+"""An error about the request must not send the model to check connectivity.
+
+2026-10-24 conversation round: a non-UUID alert_id, and a 404 that the error
+text itself explained as "this endpoint needs VCF Operations 9.0+", both came
+back with "Run 'vmware-aria doctor' to verify connectivity." Doctor reports
+green for both, so the hint pointed at the one thing that was not wrong.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import ssl
+
+import pytest
+
+from vmware_aria.config import ConfigError
+from vmware_aria.connection import AriaApiError
+from vmware_aria.mcp_server import server
+from vmware_aria.mcp_server._shared import _DOCTOR_HINT, _FIX_IN_MESSAGE_HINT, _hint_for
+
+
+@pytest.mark.parametrize("status", [400, 404, 409, 422])
+def test_request_statuses_say_the_message_has_the_fix(status: int) -> None:
+    assert _hint_for(AriaApiError("x", status_code=status)) == _FIX_IN_MESSAGE_HINT
+
+
+@pytest.mark.parametrize("status", [None, 401, 403, 500, 503])
+def test_login_server_and_transport_failures_keep_doctor(status) -> None:
+    assert _hint_for(AriaApiError("x", status_code=status)) == _DOCTOR_HINT
+
+
+@pytest.mark.parametrize("exc", [ValueError("alert_id must be a UUID"), ConfigError("set VMWARE_ARIA_X_PASSWORD")])
+def test_authored_request_errors_say_the_message_has_the_fix(exc: Exception) -> None:
+    assert _hint_for(exc) == _FIX_IN_MESSAGE_HINT
+
+
+@pytest.mark.parametrize(
+    "exc",
+    [KeyError("resourceId"), ssl.SSLError("bad cert"), json.JSONDecodeError("Expecting value", "<html>", 0), UnicodeDecodeError("utf-8", b"\xff", 0, 1, "x"), RuntimeError("boom")],
+)
+def test_errors_this_skill_did_not_write_keep_doctor(exc: Exception) -> None:
+    assert _hint_for(exc) == _DOCTOR_HINT
+
+
+def test_a_malformed_alert_id_through_the_mcp_layer(monkeypatch) -> None:
+    monkeypatch.setattr(server, "_get_connection", lambda target=None: object())
+    result = asyncio.run(server.mcp.call_tool("investigate_alert", {"alert_id": "e1fdab74-"}))
+    content = result[0] if isinstance(result, tuple) else result
+    payload = json.loads(content[0].text)
+    assert "8-4-4-4-12" in payload["error"]
+    assert payload["hint"] == _FIX_IN_MESSAGE_HINT

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/tests/test_gated_writes.py

@@ -9,7 +9,8 @@
   alert, a resource already in / not in maintenance) or a field the radius
   depends on could not be read.
 
-``confirmed`` is a deprecated alias; the conservative reading wins.
+The deprecated ``confirmed`` alias was removed (2.4.2): a call that still
+passes it is refused as an unknown argument and writes nothing.
 """
 
 from __future__ import annotations
@@ -177,36 +178,40 @@ def test_confirm_writes_once_and_returns_the_blast_radius(aria, tool):
     assert out["blast_radius"]["unmeasured"] == []
 
 
-@pytest.mark.parametrize("tool", GATED)
-def test_legacy_confirmed_true_still_acts_and_says_it_is_deprecated(aria, tool):
-    client = _ready(aria, tool)
-    out = _call(tool, confirmed=True)
-    assert len(client.writes) == 1
-    assert "confirmed is deprecated; use confirm" in out["deprecated"]
+def _protocol_call(tool: str, **kw: Any):
+    """Call through the MCP layer, where the argument model refuses unknown names."""
+    first_arg = {"acknowledge_alert": "alert_id", "cancel_alert": "alert_id",
+                 "delete_alert_definition": "definition_id", "delete_report": "report_id",
+                 "start_resource_maintenance": "resource_id",
+                 "end_resource_maintenance": "resource_id"}[tool]
+    first = {"alert_id": AID, "definition_id": DEF, "report_id": REPORT, "resource_id": RID}
+    args = {first_arg: first[first_arg], **kw}
+    if tool == "start_resource_maintenance":
+        args = {"duration_minutes": 30, **args}
+    return asyncio.run(server.mcp.call_tool(tool, args))
 
 
 @pytest.mark.parametrize("tool", GATED)
-def test_explicit_confirmed_false_holds_against_confirm(aria, tool):
+def test_the_removed_confirmed_alias_is_refused_and_writes_nothing(aria, tool):
     client = _ready(aria, tool)
-    out = _call(tool, confirm=True, confirmed=False)
-    assert out["action"] == "preview"
+    with pytest.raises(Exception, match="confirmed"):
+        _protocol_call(tool, confirmed=True)
     assert client.writes == []
-    assert "deprecated" in out
 
 
 @pytest.mark.parametrize("tool", GATED)
-def test_no_alias_no_deprecation_note(aria, tool):
+def test_no_deprecation_note_any_more(aria, tool):
     _ready(aria, tool)
     assert "deprecated" not in _call(tool)
+    assert "deprecated" not in _call(tool, confirm=True)
 
 
 @pytest.mark.parametrize("tool", GATED)
-def test_schema_defaults_confirm_false_and_alias_null(tool):
+def test_schema_defaults_confirm_false_and_has_no_alias(tool):
     t = next(x for x in asyncio.run(server.mcp.list_tools()) if x.name == tool)
     props = t.inputSchema["properties"]
     assert props["confirm"]["default"] is False
-    assert props["confirmed"]["default"] is None
-    assert "Deprecated alias for confirm" in props["confirmed"]["description"]
+    assert "confirmed" not in props
     assert "confirm" not in t.inputSchema.get("required", [])
 
 

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/tests/test_init_and_onboarding.py

@@ -115,6 +115,9 @@ def test_tls_error_hint_suggests_verify_ssl():
 
     assert _is_tls_verify_error(Exception("certificate verify failed: self signed"))
     assert not _is_tls_verify_error(Exception("Connection refused"))
+    assert not _is_tls_verify_error(
+        Exception("[SSL: UNEXPECTED_EOF_WHILE_READING] EOF occurred in violation of protocol")
+    )
 
 
 def test_tls_error_hint_offers_trusting_the_ca_before_disabling_verification():
@@ -133,3 +136,14 @@ def test_tls_error_hint_offers_trusting_the_ca_before_disabling_verification():
     assert "verify_ssl: false" in hint
     assert hint.index("SSL_CERT_FILE") < hint.index("verify_ssl: false")
     assert "lab" in hint[hint.index("SSL_CERT_FILE"):]
+
+
+def test_tls_early_eof_hint_points_to_service_or_proxy_not_ca():
+    from vmware_aria.connection import _transport_hint
+
+    hint = _transport_hint(Exception(
+        "[SSL: UNEXPECTED_EOF_WHILE_READING] EOF occurred in violation of protocol"
+    ))
+    assert "TLS connection closed unexpectedly" in hint
+    assert "Aria Operations service" in hint
+    assert "SSL_CERT_FILE" not in hint

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/tests/test_investigate.py

@@ -200,3 +200,28 @@ def test_aria_enum_values_are_passed_through_verbatim():
     assert result["alert"]["status"] == "ACTIVE"
     assert result["alert"]["alert_impact"] == "RISK"
     assert result["alert"]["control_state"] == "OPEN"
+
+
+def test_no_data_alert_hands_off_to_the_objects_that_stopped_reporting():
+    """The alert's resource is the adapter instance; the user asks which objects stopped.
+
+    2026-10-24 conversation round: next_step said "no vmware-monitor tool covers
+    VMWARE_INFRA_HEALTH_INSTANCE", which was true and left the model to find
+    list_resources(collection_status=...) on its own.
+    """
+    alert = _alert_payload()
+    alert["alertDefinitionId"] = (
+        "AlertDefinition-vCenter Operations Adapter-AdapterChildsNoDataReceivingAlert"
+    )
+    resource = _resource_payload(kind="VMWARE_INFRA_HEALTH_INSTANCE", name="Infrastructure Health Adapter")
+    resource["resourceKey"]["adapterKindKey"] = "VMWARE_INFRA_HEALTH"
+    step = investigate_alert(_client(alert=alert, resource=resource), ALERT_ID)["next_step"]
+    assert step["skill"] == "vmware-aria" and step["tool"] == "list_resources"
+    assert step["args"] == {"resource_kind": "all", "collection_status": "NO_DATA_RECEIVING"}
+    assert "VMWARE_INFRA_HEALTH" in step["note"]
+
+
+def test_other_alerts_on_an_adapter_instance_keep_the_generic_handoff():
+    resource = _resource_payload(kind="VMWARE_INFRA_HEALTH_INSTANCE", name="Infrastructure Health Adapter")
+    step = investigate_alert(_client(resource=resource), ALERT_ID)["next_step"]
+    assert step["skill"] == "vmware-monitor" and step["tool"] is None

{vmware_aria-2.4.0 → vmware_aria-2.4.2}/tests/test_policy_roles.py

@@ -53,3 +53,30 @@ def test_the_resolver_uses_the_config_the_connection_manager_holds(monkeypatch):
     held = object()
     monkeypatch.setattr(server, "_conn_mgr", types.SimpleNamespace(_config=held))
     assert pr._connection_config() is held  # the manager's copy, not the edited file
+
+
+@pytest.mark.unit
+def test_the_resolver_names_a_target_from_a_real_config(monkeypatch, tmp_path):
+    """The resolver must run against a real loaded config, not a stand-in.
+
+    It returned ``get_target_strict(...).name`` — and this skill's TargetConfig has
+    no ``name``, so it raised AttributeError on every call: a target-limited role
+    was refused everything here. No test called it with a real config (found on
+    Log-Insight, 2026-10-17; same code in five skills).
+    """
+    from vmware_aria import policy_roles as pr
+    from vmware_aria.config import load_config
+
+    conf = tmp_path / "config.yaml"
+    conf.write_text(
+        "targets:\n  lab-a:\n    host: a.example\n    username: u\n"
+        "  lab-b:\n    host: b.example\n    username: u\n"
+        "default_target: lab-b\n",
+        encoding="utf-8",
+    )
+    real = load_config(conf)
+    monkeypatch.setattr(pr, "_connection_config", lambda: real)
+    assert pr._target_name(None) == "lab-b"
+    assert pr._target_name("lab-a") == "lab-a"
+    with pytest.raises(KeyError):
+        pr._target_name("no-such-target")