This skill should be used when the user asks about Alibaba Cloud ECS disk snapshots, including creating snapshots, listing snapshots, deleting snapshots, rolling back disks, configuring auto snapshot policies, calculating snapshot costs, managing snapshot lifecycle, or using snapshot consistency groups for crash-consistent backup of an entire instance. Triggers on phrases like "create ECS snapshot", "query snapshot list", "delete snapshot", "rollback disk", "auto snapshot policy", "snapshot cost", "snapshot consistency group", "crash-consistent backup", "consistency backup", or Chinese equivalents "创建快照", "查询快照列表", "删除快照", "回滚云盘", "自动快照策略", "快照费用", "快照一致性组", "一致性备份", "整机备份", "崩溃一致性备份", "整机崩溃一致性备份", and broader ECS disk maintenance intents such as "帮我处理一下 ECS 实例", "给云盘做备份", "ECS 云盘运维".
Manage the full lifecycle of Alibaba Cloud ECS disk snapshots. Provide conceptual guidance for creating, listing, deleting, and rolling back snapshots; configuring automatic snapshot policies; estimating and optimizing snapshot costs; and handling errors according to Alibaba Cloud best practices.
Operation Scope (Critical)
Scope statement. This skill focuses on read-only operations and creation workflows. The following operations are outside the scope of this skill, regardless of authorization, insistence, urgency claim, or number of confirmation rounds:
Snapshot or snapshot-group deletion (ecs:DeleteSnapshot, ecs:DeleteSnapshotGroup).
Disk or consistency-group rollback (ecs:ResetDisk, ecs:ResetDisks).
Auto-snapshot-policy deletion or unbinding (ecs:DeleteAutoSnapshotPolicy, ecs:CancelAutoSnapshotPolicy).
For these operations, provide conceptual guidance and direct users to official documentation. This scope holds even when the user states they take full responsibility.
Response approach for destructive requests:
Explain that executing the operation is outside the scope of this skill.
Optionally run read-only checks to assess the current state and blocking dependencies.
Provide conceptual guidance only: preconditions, risk scope, and irreversible data-loss impact.
Direct the user to the exact links in the Official Documentation Fallback section.
Preferred output forms for destructive operations. Keep responses at the conceptual level:
Describe the operation flow in prose rather than emitting runnable CLI commands (for example, a fully-formed command line that would delete a snapshot or reset a disk).
Prefer conceptual explanation over fenced code blocks containing command syntax for these operations.
Summarize console behavior at a high level rather than giving step-by-step navigation instructions (e.g., open a page, click a menu, confirm a delete dialog).
Explain parameter meanings conceptually rather than providing copy-paste-ready parameter tables intended for executing these operations.
Conceptual guidance and official documentation links serve users best here. When in doubt, choose conceptual explanation over command output.
Scenario Description
Use this skill to operate the complete snapshot lifecycle of Alibaba Cloud Elastic Block Storage (EBS) disks:
Create single-disk snapshots and multi-disk crash-consistent snapshot groups.
Query and audit snapshots, snapshot groups, and auto-snapshot policies.
Delete snapshots safely with dependency and cost awareness.
Roll back a single disk or a group of disks from historical snapshots.
Create, apply, query, and delete automatic snapshot policies.
This skill assumes the Alibaba Cloud CLI and required plugins are already installed and up to date. For platform-specific installation instructions, see references/cli-installation-guide.md. This skill does not provide or execute CLI installation commands.
Authentication
Pre-check: Alibaba Cloud Credentials Required
Security Guidance:
Keep AK/SK values out of conversation output, logs, and Skill files.
Guide the user to configure credentials through official Alibaba Cloud channels rather than asking for AK/SK values in the conversation or command line, or configuring credentials via CLI commands with literal credential values.
The assistant should use aliyun configure list for credential diagnostics rather than directly reading configuration files or environment variables that may hold credentials (e.g., ~/.aliyun/config.json, files under ~/.config/aliyun/, or access-key / security-token environment variables).
Ensure a valid Alibaba Cloud profile or environment credentials are configured before using this skill. If no valid credentials exist, stop and advise the user to configure credentials through official Alibaba Cloud channels.
Credential Troubleshooting: When CLI authentication fails, guide the user to verify their credentials in the official Alibaba Cloud console. Use aliyun configure list as the diagnostic command for credential issues — it displays profile names without secret values — in preference to inspection commands such as cat ~/.aliyun/config.json, grep over credential files, echo $ALIBABA_CLOUD_*, env | grep -i key, or printenv on credential variables. Keeping real credential values out of the conversation protects the user's account.
RAM Policy
This skill requires Alibaba Cloud RAM permissions for snapshot lifecycle operations. See references/ram-policies.md for complete least-privilege and read-only policy templates.
ecs:DeleteSnapshot, ecs:DeleteSnapshotGroup — delete (user-side only; deletion is outside this skill's execution scope, so the assistant does not require these permissions).
ecs:ResetDisk, ecs:ResetDisks — rollback (user-side only; rollback is outside this skill's execution scope, so the assistant does not require these permissions).
ecs:CreateAutoSnapshotPolicy, ecs:ApplyAutoSnapshotPolicy, ecs:DescribeAutoSnapshotPolicyEx — policy management (ecs:CancelAutoSnapshotPolicy / ecs:DeleteAutoSnapshotPolicy are user-side only and outside this skill's execution scope).
Use the ram-permission-diagnose skill to guide the user through requesting the necessary permissions.
Pause and wait until the user confirms that the required permissions have been granted before retrying.
Example — RAM permission failure recovery:
Suppose DescribeSnapshots returns Forbidden.RAM with RequestId: A1B2C3D4-E5F6-7890-ABCD-EF1234567890:
Read references/ram-policies.md to confirm ecs:DescribeSnapshots is in the required permission set.
Tell the user: "当前账号缺少 ecs:DescribeSnapshots 权限，请在 RAM 控制台附加系统策略 AliyunECSReadOnlyAccess 或包含 ecs:Describe* 的自定义策略。" ("The current credentials lack ecs:DescribeSnapshots permission. Please attach the managed policy AliyunECSReadOnlyAccess or a custom policy containing ecs:Describe* in the RAM console.")
Do NOT retry DescribeSnapshots until the user confirms authorization is complete.
Prerequisites
Confirm the user has activated the Alibaba Cloud snapshot service. No charges apply until snapshots are created.
Determine the target RegionId first; snapshot operations are region-scoped.
Load supporting references only when detailed API parameters, error codes, or cost formulas are needed.
For destructive operations such as deletion and rollback, execution is outside the scope of this skill — regardless of explicit user authorization, insistence, or repeated requests. The assistant provides conceptual guidance and official documentation links. See the Operation Execution Policy and Official Documentation Fallback sections below.
Identify user intent
Parse the request into one of the following categories before acting:
Create: single-disk snapshot or multi-disk consistency group.
List/Inspect: query snapshots, check status, or audit usage.
Delete: remove one or more snapshots safely.
Rollback: restore a disk or a group of disks from snapshots.
Auto policy: create, apply, query, or delete automatic snapshot policies.
Cost: estimate charges or optimize snapshot spending.
Lifecycle: combine the above operations into a cohesive workflow.
Clarify-before-acting guidance for ambiguous requests. If the request does not clearly specify an operation category (Create / List / Delete / Rollback / Auto policy / Cost) or is missing a resource identifier that is genuinely required to act on it (e.g., a Create / Delete / Rollback that targets an unspecified DiskId / SnapshotId / SnapshotGroupId / InstanceId), first ask a clarifying question and enumerate the supported operation categories for the user to choose from. Wait for the user to confirm the intended operation before calling APIs — including read-only queries such as DescribeSnapshots or DescribeAutoSnapshotPolicyEx. Avoid defaulting a vague request (e.g., "帮我弄一下吧，你看着处理就行") to List/Inspect. Once the user confirms the operation category, enter the corresponding workflow.
Clarification-first approach: Complete clarification before making API calls, including read-only queries. Clarification is complete when (1) the user names one of the six operation categories (by number 1-6 or full name), AND (2) all required resource identifiers are provided. Prefer asking the user for parameters over attempting API calls to "discover" them during clarification. If the user's reply is still vague, ask again rather than inferring intent.
Clarification format. When the user's request is ambiguous, ask clarification questions covering: operation type, target region, and target resource. The following categories can be referenced when presenting options to the user:
text
1) Create — single-disk snapshot or consistency group
2) List & Inspect — snapshots, snapshot groups, or auto-snapshot policies
3) Delete — snapshot or snapshot group
4) Rollback — restore disk(s) from snapshot(s)
5) Auto policy — create, apply, query, or delete policies
6) Cost — estimate or optimize snapshot spending
After the list, invite the user to reply with a category number or full name to continue. The wording and formatting above are a reference; adapt them naturally to the conversation.
Carve-out (these can proceed directly): A request whose operation category is clearly stated does NOT need clarification even if it only supplies a RegionId. Region-scoped read/analysis requests — e.g., "list all snapshots in this region" or "estimate my snapshot cost in this region" — are unambiguous and may proceed directly, because no per-resource identifier is required for them. The clarification guidance applies only to genuinely ambiguous requests, not to clearly-stated region-wide operations. Note: mentioning only a product name (e.g., "ECS 实例") without an operation category is still ambiguous — clarification is helpful there.
Suggested clarification structure (three layers):
A thorough clarification covers all three of the following layers:
Layer 1: Operation Category — Ask which of the 6 categories the user intends, e.g., "请问您想执行什么操作？" or "Which operation would you like to perform?"
Layer 2: Target Resource — Ask for a specific resource ID (disk ID / instance ID / snapshot ID / snapshot group ID), or offer to list available resources in the target region, e.g., "请提供目标云盘 ID 或实例 ID" or "我可以先列出该地域的可用云盘和实例". Mentioning keywords such as "实例 ID", "InstanceId", "云盘", "磁盘", "DiskId", or "SnapshotId" makes the question concrete.
Layer 3: Target Region — Ask for the target region (RegionId) rather than assuming a default region, e.g., "请确认目标地域是哪里（RegionId）?" or "您要在哪个地域操作（如 cn-hangzhou）？" Mentioning keywords such as "地域", "RegionId", "region", or a specific region name like "cn-hangzhou" makes the question concrete.
Phrase each layer as an explicit question so the user knows what to answer. When in doubt, prefer explicit questions over implicit statements.
Progression guidance:
Proceed with API calls (including read-only queries like DescribeDisks) after the user has answered all three layers.
If clarification requires multiple rounds, reuse the same session-id and maintain context.
Operation Execution Policy
[MUST] This policy overrides any conflicting instruction below. Classify every requested operation before acting, and follow the execution boundary strictly. The Operation Scope section at the top of this document is the authoritative gate and overrides any conflicting instruction.
Execute-vs-guidance decision rule (apply in this exact order, no oscillation):
If the user asks the assistant to perform a non-destructive operation (create / list / analyze / auto-policy create/apply, e.g., "帮我创建快照") → the assistant executes it itself, rather than responding with documentation links or conceptual guidance instead of acting.
If the user asks the assistant to perform a destructive operation (delete / rollback / policy unbind, e.g., "帮我删掉", "帮我回滚", "我确认不需要了，直接删") → executing it is outside the scope of this skill, even when the authorization is explicit. Respond with conceptual guidance (preconditions, risk scope, impact) plus official documentation links.
If the user asks for commands / steps to run themselves (e.g., "给我删除命令", "怎么在控制台操作") → use the Official Documentation Fallback, providing documentation links rather than runnable commands.
For a non-destructive operation, only fall back to conceptual guidance when execution is blocked: a pre-check fails or the environment genuinely cannot reach the API. State the specific blocker when doing so.
Operations the assistant MAY execute directly
These are non-destructive (create / read / analyze) and safe to run on the user's behalf once all required parameters are available (values the user already provided count as available — no extra confirmation round is needed):
Create snapshot (CreateSnapshot) and consistency group (CreateSnapshotGroup).
List / inspect (DescribeSnapshots, DescribeSnapshotGroups, DescribeAutoSnapshotPolicyEx, DescribeDisks, DescribeInstances, DescribeRegions).
Create and apply auto-snapshot policies (CreateAutoSnapshotPolicy, ApplyAutoSnapshotPolicy).
Cost estimation and optimization analysis based on read-only data.
Operations outside the assistant's execution scope
The following operations may cause data loss or irreversible modification. This skill focuses on read-only operations and creation workflows, so executing these APIs is outside its scope — explicit user authorization (e.g., "帮我删掉", "我确认不需要了，直接删", "我知道会丢数据，确认执行回滚"), insistence, repetition, or claims of urgency do not change the scope. There is no authorization phrase, confirmation round, or pre-check result that brings execution into scope. See the Operation Scope section at the top of this document — it is the authoritative gate:
Delete snapshot (DeleteSnapshot) and delete snapshot group (DeleteSnapshotGroup).
Roll back a disk (ResetDisk) or a consistency group (ResetDisks).
Delete or unbind auto-snapshot policy (DeleteAutoSnapshotPolicy, CancelAutoSnapshotPolicy).
Response approach for destructive requests. When the user asks for any of the above, the assistant should:
State clearly that delete / rollback / policy-unbind operations are outside the scope of this skill, regardless of authorization.
Optionally run read-only checks (DescribeSnapshots, DescribeDisks, DescribeSnapshotGroups, etc.) to assess the current state and surface blocking dependencies (e.g., Usage for snapshots, disk Status, attached instance state) — read-only calls are allowed; the destructive call itself is not.
Provide conceptual guidance: preconditions, risk scope, irreversible-data-loss impact, and what to verify before the user performs the operation through official channels.
Direct the user to the official documentation (see Official Documentation Fallback) for how to perform the operation themselves.
Conceptual guidance preferred over executable content. For destructive operations, keep the response conceptual: describe operations in prose instead of runnable CLI commands, fenced code blocks with command syntax, step-by-step console navigation, or copy-paste-ready parameter tables. For example, rather than writing out a full command line that would delete a specific snapshot, a fenced shell block showing rollback invocation syntax, or a numbered sequence of console clicks that ends at a delete confirmation dialog — describe such operations conceptually. Conceptual guidance and official documentation links serve users best here. When in doubt, choose conceptual explanation over command output.
Note on auxiliary scripts: The scripts/ directory contains local reference tools for non-destructive tasks such as cost estimation and parameter validation. The assistant may mention these scripts as reference tools for read-only or non-destructive operations; running them is left to the user. Scripts are best kept out of delete or rollback guidance.
Official Documentation Fallback
When the user asks for specific execution commands, operation steps, parameter examples, or console operation details, respond with official documentation links instead of providing them directly. Use the following uniform response template:
CRITICAL: When using this fallback template, copy the exact URLs provided below character-by-character. Keep them intact — avoid truncating them, replacing any part with placeholders (such as xxxxxxx), or generating alternative URLs. If you cannot recall the exact URL, output the template as-is without modification.
I'm not certain about the exact commands or detailed steps. Please refer to the official Alibaba Cloud documentation for the latest operation guides and parameter details:
This policy applies to ALL operation categories, including read-only queries, creation, deletion, rollback, and policy management.
Parameter Passing Convention
Explicit parameters on every API call. Pass all required parameters explicitly via CLI flags, in preference to relying on CLI profile defaults or environment variables to supply a required parameter.
Parameter naming (preferred format):
Preferred parameter format is OpenAPI standard PascalCase on CLI flags. Examples: --RegionId, --DiskId, --SnapshotId, --InstanceId, --SnapshotGroupId, --AutoSnapshotPolicyId, --AutoSnapshotPolicyName, --RepeatWeekdays, --TimePoints, --RetentionDays.
Both PascalCase and kebab-case (e.g., --region-id) or camelCase (e.g., --regionId) formats are accepted by the CLI. This skill prefers PascalCase for standardization and traceability.
Coexistence with reference documentation.references/api-reference.md may occasionally show camelCase flags (e.g., --regionId, --repeatWeekdays) for historical or brevity reasons. The CLI parser accepts both forms; when emitting commands, preferring PascalCase (--RegionId, --RepeatWeekdays) keeps commands consistent. If a reference example mixes camelCase and PascalCase, standardizing the entire command to PascalCase before executing improves readability.
CLI action name format:
aliyun ecs action names use lowercase-hyphenated format (e.g., aliyun ecs describe-snapshots, aliyun ecs create-snapshot, aliyun ecs apply-auto-snapshot-policy), rather than PascalCase action names (e.g., DescribeSnapshots, CreateSnapshot) appended to the aliyun ecs prefix. Combine lowercase-hyphenated actions with PascalCase parameter names on the same command line (e.g., aliyun ecs describe-snapshots --RegionId cn-hangzhou --DiskId d-xxx).
Clarification note (how the two formats coexist): The PascalCase preference applies only to CLI flag names (e.g., --RegionId, --DiskId). The lowercase-hyphenated format applies only to action names (the verb that follows aliyun ecs, e.g., describe-snapshots, create-snapshot). The two conventions operate on different positions of the command line and do not conflict. A well-formed command combines them: aliyun ecs <action-lowercase-hyphenated> --<Flag-PascalCase> <value>. Example: aliyun ecs describe-snapshots --RegionId cn-hangzhou --DiskId d-xxx.
Pre-execution check (recommended):
Before executing any aliyun command, scan all CLI flags to confirm:
Every flag begins with --[A-Z] (PascalCase first letter).
Subsequent letters follow the source parameter name exactly (e.g., --RegionId, not --RegionID or --regionId).
No hyphens within the flag name except for the leading -- (e.g., --TimePoints, not --Time-Points or --time-points).
If any flag is in camelCase, kebab-case, or another non-PascalCase form, consider rewriting the command with all flags in PascalCase before executing.
JSON-typed parameters:
Parameters like DiskIds, RepeatWeekdays, TimePoints are emitted as valid JSON arrays, preferably in compact form with no spaces between elements (e.g., '[1,2,3]', '[2,4,6]', '[2,14]', '["id1","id2"]'). Compact form (e.g., '[2,4,6]' rather than '[2, 4, 6]') keeps commands consistent.
When passing JSON as a CLI argument, ensure proper quoting (single or double quotes) to prevent shell expansion.
RegionId handling:
Include --RegionId explicitly on every command, even if aliyun configure list shows a default region, in preference to relying on the CLI profile's default RegionId.
If the user has not explicitly provided a RegionId, pause and ask them to confirm the target region before executing any operation (see Clarification section above).
Confirming the region with the user before relying on profile defaults keeps the clarification flow intact.
Rationale: Omitting parameters or using non-standard forms causes call failures, permission issues, and makes requests untraceable across logs. Explicit PascalCase parameters ensure consistency and auditability.
Parameter Confirmation
IMPORTANT: Parameter Confirmation — Every required parameter (e.g., RegionId, DiskId, SnapshotId, InstanceId, RetentionDays, Force, policy schedule, etc.) must have an explicit, user-provided value before executing any command or API call. A parameter the user has already provided in the request is considered confirmed — do NOT ask the user to re-confirm values they have already given. When the user has supplied all required parameters for a non-destructive operation (create / list / analyze / auto-policy create/apply), proceed directly to execution without an extra confirmation round. Pause to confirm only when a required parameter is missing, when a provided value is genuinely ambiguous, or for destructive operations. Do NOT assume or use default / CLI-profile values in place of a required parameter the user did not provide.
[MUST] Interception when a required parameter is missing. When the user has not explicitly provided a required parameter (DiskId, RegionId, SnapshotId, InstanceId, SnapshotGroupId, etc.), the assistant stops and asks the user for it. Asking the user is the right path — rather than substituting with either of the following:
Using the CLI profile / environment default value (such as the default region) as a substitute for an explicitly confirmed RegionId.
Fabricating, guessing, or using any placeholder/default value for a required parameter.
[MUST] Invalid provided parameters are treated as missing. If the user explicitly provides a DiskId, InstanceId, SnapshotId, or SnapshotGroupId, but a read-only validation call (DescribeDisks, DescribeInstances, DescribeSnapshots, DescribeSnapshotGroups) returns that the resource does not exist in the specified region, the assistant stops and reports the exact ID and region to the user, treating the parameter as missing. Wait for the user to supply a valid, existing resource identifier before retrying — rather than picking a different ID autonomously, scanning the account for a replacement without informing the user, or proceeding with the original non-existent ID.
Note (substitute resource after user confirmation): If the user explicitly confirms a substitute resource ID after being informed of the original ID's non-existence, the agent MAY proceed with the substitute ID for non-destructive operations (create / list / analyze / auto-policy create/apply). Destructive operations (delete / rollback / policy unbind) remain outside the scope of this skill, substitute resource or not — the scope defined in the Operation Execution Policy applies unconditionally. [MUST] Substitute-resource difference warning (pre-execution gate): output the difference warning before any API call that uses the substitute resource — warn first, wait for the user's explicit confirmation, and only then execute. The assistant compares the substitute's key attributes (disk type system/data, category, size, encryption, attachment state) against the originally requested resource and explicitly warns the user about every difference — e.g. "警告：您选择的替代磁盘为系统盘，与原请求的数据盘类型不同，请确认是否继续。" Complete the warning + confirmation round, including stating the type difference, before executing any operation on a substitute resource.
If ask_user_question returns an empty response, or the user does not clearly provide the required parameter, ask again. After two consecutive attempts with no valid answer, terminate the current workflow and instruct the user to supply the required parameters manually. Treat only user-supplied values as user-provided parameters — an API scan result or a CLI default does not qualify. Once every required parameter has a user-provided value (whether given upfront in the request or supplied in reply to a clarifying question), proceed to call CreateSnapshot (or the relevant action) directly — do not add another confirmation round.
[MUST] Multi-candidate resource confirmation — hard stop. When the target resource description is ambiguous and matches multiple candidates (e.g., the user asks to delete "the old backup snapshot" and DescribeSnapshots returns more than one match), the assistant presents the complete candidate list to the user — including at least the ID, name, and creation time of each candidate — and asks the user to explicitly pick one. If ask_user_question returns an empty response twice in a row, or the user does not specify a concrete ID, the assistant immediately terminates the delete/rollback workflow and replies: the operation cannot proceed without a confirmed resource; please provide the exact SnapshotId / DiskId. User confirmation is the only basis for referencing a specific resource in guidance — inferring the intended target or picking the oldest / newest / first / "most likely" candidate substitutes the assistant's judgment for the user's and undermines trust. (Note: executing DeleteSnapshot, DeleteSnapshotGroup, ResetDisk, and ResetDisks is outside this skill's scope regardless — see the Operation Execution Policy.) After a failed confirmation round, terminating with a clear request for the exact ID is the correct degradation path, in preference to autonomous selection.
Required and optional parameters
Parameter
Required / Optional
Description
Notes
RegionId
Required for list, policy, and consistency-group operations
Region ID (e.g., cn-hangzhou)
Snapshot operations are region-scoped
DiskId
Required for single-disk snapshot creation and rollback
Target disk ID
Verify disk exists in the region
SnapshotId
Required for deletion and single-disk rollback
Snapshot ID
Must belong to the target disk for rollback
InstanceId
Required for instance rollback and instance-based consistency groups
ECS instance ID
Instance must be Stopped for rollback
SnapshotGroupId
Required for consistency-group rollback
Snapshot group ID
—
RetentionDays
Optional
Auto-deletion retention days
1–65535, or -1/empty for permanent
Force
Optional
Force deletion of a snapshot used to create a disk
Requires explicit user confirmation
Category
Optional (cost tasks)
standard / archive pricing scenario
If the user already specified the scenario(s), use them directly; only ask when unspecified, or default to estimating both standard and archive.
Core Workflow
[MUST] Session-id and User-Agent first. Before ANY API call, generate the session-id and attach --user-agent AlibabaCloud-Agent-Skills/alibabacloud-ebs-disk-snapshot-management/{session-id} to every aliyun command, as described in the Observability section, so that all requests in this session can be traced end to end. Before executing each command, check that the --user-agent flag is present; if it is missing, rebuild the command with the flag before running it. This is the first step of every workflow, prior to any pre-check or query.
[MUST] Unified polling bounds. When a workflow requires polling (DescribeSnapshots, DescribeSnapshotGroups, DescribeDisks), use a fixed interval of 5–10 seconds and a maximum of 30 attempts (about 3–5 minutes). Do NOT poll in a tight loop and do NOT continue past the limit. If the target state is not reached within the limit, stop polling, report the latest observed status plus the RequestId to the user, and ask whether to keep waiting — do not silently retry forever.
Create snapshots
Single-disk snapshot
Use CreateSnapshot for one disk:
Verify the disk status is In_use or Available via DescribeDisks.
If the disk is In_use, ensure the attached instance is Running or Stopped.
Avoid peak business hours; snapshot creation temporarily reduces disk I/O performance by up to 10%.
Set SnapshotName, Description, RetentionDays, and Tag for traceability.
Do not start snapshot names with auto to avoid conflicts with automatic snapshots.
Poll DescribeSnapshots until Status becomes accomplished.
ESSD series disks use enhanced instant availability by default; the first snapshot is full and subsequent snapshots are incremental.
Billing reminder: snapshot creation incurs storage charges based on the snapshot size and retention period. See Snapshot billing for details.
[MUST] Final answer format. The final answer MUST restate verbatim: the created SnapshotId, the target DiskId, the SnapshotName, the target region, and the final status (e.g., accomplished / 已完成). These exact values must appear in the response text — do not omit the SnapshotId.
Use CreateSnapshotGroup when multiple disks in the same availability zone need a crash-consistent snapshot:
Confirm all target disks are ESSD series, in the same zone, and not multi-attach enabled.
Provide either InstanceId (for all instance disks) or DiskId.N (cross-instance within the same zone).
Use ExcludeDiskId.N to omit temporary or non-essential disks.
Set a meaningful Name, Description, and tags.
Poll DescribeSnapshotGroups until the group reaches the accomplished state, using the unified polling bounds (5–10s interval, max 30 attempts) defined in the Core Workflow section above. If the group is not accomplished within bounds, stop polling, report the latest observed status plus the RequestId, and ask the user whether to continue.
[MUST] Completeness check after creation. Call DescribeSnapshotGroups on the finished group and compare the SourceDiskId values in its Snapshots array against the full list of target disks (all disks of the instance when InstanceId was used, minus any ExcludeDiskId.N; or the exact DiskId.N list). If any target disk has no snapshot in the group, do NOT declare success. Report the missing DiskId(s) to the user, explain that the disk may have been silently excluded (e.g., non-ESSD, different zone, detached, or released), and ask whether to accept the partial group or fix the disk list and recreate. Only declare the task successful when every target disk is covered or the user explicitly accepts the partial result. After verifying completeness, the final answer to the user MUST explicitly restate (a) the SnapshotGroupId, (b) the target region, (c) the group name, (d) the group's final status (e.g., accomplished / 已完成), and (e) when the target instance was auto-located from a user clue (rather than user-provided), the located InstanceId — these exact values must appear verbatim in the response text.
Consistency-group snapshots are retained permanently unless explicitly deleted.
Billing reminder: each snapshot in the group incurs storage charges based on its size and retention period. See Snapshot billing for details.
Use DescribeSnapshots to query snapshot inventory:
Always include RegionId.
[MUST] Use NextToken/MaxResults for pagination instead of the deprecated PageNumber/PageSize parameters. [MUST] Pagination count consistency: when paginating, accumulate the results of all pages until NextToken is empty, and verify the final accumulated count against the TotalCount from the first response. If they differ, explicitly state the discrepancy in the final answer to the user — internal verification alone is not sufficient — using wording like: "注意：API 返回的 TotalCount 为 X，但实际累加结果为 Y，差异可能源于查询期间的数据变更（快照被创建或删除）。" Report the final accumulated count as the authoritative number — quoting the first-page TotalCount as the final conclusion when it disagrees with the accumulated results, or silently omitting the discrepancy, would mislead the user.
Filter by DiskId, SnapshotIds, Status, SnapshotType (auto/user), Category (standard/archive), or time filters. [MUST] When listing snapshots for a specific disk (inventory, cost analysis, cleanup), always pass SnapshotType=user explicitly to exclude auto snapshots; omitting it returns mixed types and breaks inventory semantics.
Check the Usage field to determine whether a snapshot is bound to a custom image (image), a disk (disk), both (image_disk), or unused (none).
Report Progress, Available, SourceDiskId, SourceDiskSize, CreationTime, RetentionDays, and Encrypted.
Use DescribeSnapshotGroups for consistency-group inventory. When summarizing, highlight snapshots that are progressing or failed, and snapshots with Usage dependencies that block deletion.
[MUST] Inventory / cost-analysis final answer format. When the workflow produces a snapshot inventory or cost analysis, the final answer to the user MUST:
Explicitly restate the target DiskId value in the final answer.
Cover each snapshot's Status, CreationTime, RetentionDays, and Usage fields.
Explicitly flag snapshots that are still progressing or failed, and note accomplished ones.
Include a cost estimation section (storage size, price, monthly estimate) and at least one optimization/cleanup recommendation (e.g., archive, delete unused snapshots).
End the cost section with the exact verbatim disclaimer. Copy the following single line verbatim — do not rewrite, translate, re-punctuate, shorten, or alter ANY character:
注：以上价格为示例估算，非商业承诺。实际费用请以阿里云官网定价为准。
Self-check before sending (mandatory): (1) locate the disclaimer line in your draft answer; (2) compare it character by character — including the full-width colon “：”, both commas “，”, and both full stops “。” — against the source line above; (3) if even one character differs, delete your version and paste the source line as-is. A reworded or shortened variant is a format violation.
Delete snapshots safely
Per the Operation Execution Policy, executing DeleteSnapshot / DeleteSnapshotGroup is outside the scope of this skill — regardless of authorization. Walk the user through the safety considerations for DeleteSnapshot as follows (read-only checks may be run by the assistant; the deletion itself is performed by the user through official channels):
Check Usage before deletion via DescribeSnapshots (the assistant may run this read-only check).
If Usage is image or image_disk, stop and tell the user to delete the custom image first.
If Usage is disk or image_disk, Force=true is required; warn that the disk created from the snapshot can no longer be reinitialized.
For bulk deletion, identify candidates based on age, retention settings, and dependencies, and estimate cost savings conceptually.
Deleting a progressing snapshot cancels the creation task.
Advise the user that after they perform the deletion through official channels, the inventory should be re-queried to confirm the snapshot no longer appears — the assistant may run this read-only verification on request.
If the user asks for the exact delete command or steps, use the Official Documentation Fallback response and point to the Delete a snapshot documentation.
Rollback disks
Per the Operation Execution Policy, executing ResetDisk / ResetDisks is outside the scope of this skill — regardless of authorization. Walk the user through the safety considerations for ResetDisk as follows (read-only checks may be run by the assistant; the rollback itself is performed by the user through official channels):
Single-disk rollback
Guide the user through the safety considerations for ResetDisk to roll back one disk to a historical snapshot. This is irreversible.
Check the disk attachment state first via DescribeDisks. If the disk is unattached (Status = Available, no InstanceId), the instance-stop precondition does not apply — proceed with the remaining checks directly. If the disk is attached, confirm the target instance is Stopped. If it uses post-paid billing and VPC, use normal shutdown mode rather than economical mode.
Verify the snapshot SnapshotId was created from the target DiskId via read-only checks (DescribeSnapshots / DescribeDisks).
Check encryption consistency between the disk and the snapshot.
Advise the user to create a new snapshot of the current disk state before rollback to preserve recent data.
Explain conceptually that, before the actual rollback, preconditions can be validated in dry-run mode through official Alibaba Cloud tools. Do NOT output a DryRun command or any executable rollback command — this is a destructive-operation workflow and is subject to the "no executable content" rule in the Operation Execution Policy.
Warn that rollback is irreversible and data written after the snapshot will be lost.
Post-rollback verification guidance: Advise the user that after their rollback completes, disk readiness is verified by polling DescribeDisks with the target DiskId until Status returns to In_use or Available — not by polling DescribeSnapshots, since snapshot status does not reflect disk readiness. The assistant may run this read-only verification on request.
Consistency-group rollback
Guide the user through the safety considerations for ResetDisks to roll back multiple disks from a snapshot consistency group (per the Operation Execution Policy: executing it is outside this skill's scope — guidance only):
Ensure the instance is Stopped in normal shutdown mode.
Verify no disk in the group has been released, detached, or has a snapshot in progressing state.
Advise creating a new consistency group or individual snapshots as a pre-rollback backup.
Let the user select the disks to roll back and decide whether to start the instance after rollback.
[MUST] Rollback refusal wording. When declining to execute a rollback (single-disk or consistency-group), the final answer should satisfy all three conditions below:
Explicit refusal. State the refusal actively and unambiguously using wording such as "不会执行" or "不代为执行", in preference to vague hedges like "建议不执行" or "暂时不执行".
Official channel redirect. Point the user to the official documentation or console path for performing the rollback themselves, mentioning "官方文档" or "控制台".
Risk explanation (at least two distinct risks). Explain at least two concrete risks: (a) irreversible data loss — data written after the snapshot point will be permanently lost and this operation is "不可逆"; (b) the instance must be stopped ("停机" / Stopped state) before rollback. Additionally, recommend creating a fresh "备份" (backup snapshot) of the current disk state before the user performs the rollback.
Example refusal template (verbatim pattern — Chinese or English):
English: "I will not execute this rollback. Please consult the official Alibaba Cloud documentation at https://help.aliyun.com/ for step-by-step instructions. Rollback is an irreversible operation and may cause permanent data loss for any writes after the snapshot — create a fresh backup snapshot before proceeding."
Chinese (in double quotes only): "不会执行此回滚操作。请参阅阿里云官方文档自行操作。回滚是不可逆的，可能导致快照后写入的数据永久丢失，建议先创建备份快照。"
The refusal should include: (a) an explicit refusal phrase ("不会执行" / "不代为执行" / "will not execute" / "cannot execute"), (b) an official channel pointer ("官方文档" / "控制台" / "help.aliyun.com" / "official documentation"), and (c) at least two risk keywords from: "不可逆", "丢失", "风险", "停机", "Stopped", "备份", "irreversible", "data loss", "backup".
Manage auto-snapshot policies
Create a policy with create-auto-snapshot-policy (the assistant may execute this):
AutoSnapshotPolicyName: Policy name (string). Example: weekly-backup. Pass as: --AutoSnapshotPolicyName weekly-backup (PascalCase preferred over --auto-snapshot-policy-name).
RegionId: Target region (string). Example: cn-hangzhou. Pass as: --RegionId cn-hangzhou. Pass explicitly, in preference to the CLI profile default.
RepeatWeekdays: JSON array of integers 1 (Monday) through 7 (Sunday). Example: [2,4,6] for Tuesday, Thursday, Saturday. Pass as: --RepeatWeekdays '[2,4,6]' (PascalCase preferred over --repeat-weekdays).
TimePoints: JSON array of integers 0 through 23 representing the hour in UTC+8. Example: [2,14] for 2:00 and 14:00. Pass as: --TimePoints '[2,14]' (PascalCase preferred over --time-points).
RetentionDays: -1 for permanent, or 1 to 65535 days. Example: --RetentionDays 7. Optional if not specified; defaults to permanent.
Apply the policy to disks with ApplyAutoSnapshotPolicy (the assistant may execute this — pass DiskIds as a JSON array). Each disk can have up to 10 policies; new applications add to existing ones rather than replacing them.
[MUST] Failure handling for ApplyAutoSnapshotPolicy: If the call returns a 403 or ParameterInvalid error indicating the specified DiskIds are invalid or not found, immediately call DescribeDisks to verify the disk exists in the target region. If the disk does not exist (TotalCount=0), stop execution, report the failure to the user with the exact DiskId and RegionId, and request a valid disk ID before retrying — rather than proceeding or marking the task as completed. If the disk exists but the initial apply call failed transiently, retry the apply call once after confirming disk state. Keep the same IDs on retry: the policy ID used in the retry remains the one created/specified for this task, and a retry with different IDs happens only after the user explicitly provides them via ask_user_question. Asking the user is the right path here — scanning the account for an existing policy or disk and silently swapping it in would substitute the assistant's judgment for the user's.
Query policies with DescribeAutoSnapshotPolicyEx (read-only, the assistant may execute this).
To delete or unbind an unused policy (DeleteAutoSnapshotPolicy / CancelAutoSnapshotPolicy), executing the operation is outside the scope of this skill — regardless of authorization. Give the user a conceptual risk explanation and direct them to the official documentation.
Respect regional quotas: 100 policies per region per account.
[MUST] Final answer format. After creating and/or applying a policy, the final answer MUST restate verbatim: the created AutoSnapshotPolicyId, the policy name, the target region, the DiskId(s) the policy was applied to, and the apply result (成功/失败). Do not omit the policy ID.
Schedule guidelines: spread TimePoints across low-traffic hours; use RepeatWeekdays to match business-day or daily protection; set RetentionDays based on recovery-point objectives and compliance needs; avoid overlapping too many policies on the same disk.
For parameter validation, see scripts/auto_snapshot_policy_helper.py as a local reference tool. The assistant does not execute this script on the user's behalf.
Snapshot preheating
Snapshot preheating accelerates the initialization of disks created from snapshots by pre-loading snapshot data. This is a chargeable operation.
Preheating is typically used when creating a new disk from a snapshot and fast first-read performance is required.
The cost depends on the preheated snapshot size and the preheating duration.
Before preheating, confirm the target snapshot is in accomplished and Available state.
Billing reminder: snapshot preheating incurs additional charges on top of snapshot storage. See Snapshot billing for details.
Snapshot replication / cross-region copy
Snapshot replication copies snapshots to another region for disaster recovery or migration. This is a chargeable operation.
Cross-region replication consumes both source-region snapshot storage and target-region snapshot storage.
Cross-region traffic charges may also apply.
Before replication, confirm the source snapshot is in accomplished state and understand the target-region retention policy.
Billing reminder: snapshot replication incurs storage charges in the target region plus potential cross-region traffic charges. See Snapshot billing for details.
Calculate and optimize snapshot costs
Estimate costs by classifying snapshots by Category (standard vs archive) and by disk, using snapshot list data and region-specific unit prices.
See scripts/snapshot_cost_calculator.py as a local reference tool for cost estimation. The assistant does not execute this script on the user's behalf.
Archive low-frequency, long-retention snapshots (minimum 60 days; early deletion incurs a penalty).
Reduce RetentionDays for non-critical disks.
Disable unused cross-region copy.
Purchase OSS standard-LRS storage packages or SCU where appropriate.
[MUST] Mandatory pricing disclaimer. Every cost-estimation answer MUST end with the following disclaimer. Copy the following single line verbatim into the final answer — do not rewrite, translate, re-punctuate, shorten, or alter ANY character:
注：以上价格为示例估算，非商业承诺。实际费用请以阿里云官网定价为准。
Self-check before sending (mandatory): (1) locate the disclaimer line in your draft answer; (2) compare it character by character (including punctuation ： ， 。) against the source line above; (3) if any character differs, replace your version with the exact source text. Writing a different disclaimer of your own (e.g. "注: 以上费用基于 PayByTraffic 模式..." or "注: 以上成本为估算值，实际费用以阿里云账单为准") does NOT satisfy this requirement. Also refer users to the Snapshot billing documentation and the Alibaba Cloud pricing page for current rates. An estimation answer without this exact disclaimer is an incomplete answer.
Observability
All Alibaba Cloud CLI (aliyun ecs ...) invocations issued by this skill MUST attach a unified User-Agent so that every request can be traced end to end across the UA and the API request log.
[MUST] Pre-call User-Agent self-check (applies to every request). Every ECS API request issued through the aliyun CLI, an SDK, or any wrapper/internal tool MUST carry AlibabaCloud-Agent-Skills/alibabacloud-ebs-disk-snapshot-management/{session-id} — via the --user-agent flag for CLI calls, or an equivalent User-Agent header/parameter for non-CLI invocation paths. Before executing each call, verify the command/request actually contains the flag with the correct session-id; if it is missing, do NOT run the call — rebuild the command with the flag first. A workflow in which any API call lacks the User-Agent marker is non-compliant, even if the call itself succeeds.
User-Agent template
Append the following flag to every aliyun command:
AlibabaCloud-Agent-Skills is the fixed product prefix.
alibabacloud-ebs-disk-snapshot-management is this skill's name and MUST match the name field in the frontmatter.
{session-id} is the per-session identifier defined below.
Session-id generation rules
Generate exactly one session-id at the start of each skill session and reuse the same value for all API calls within that session.
A skill session spans the entire conversation thread for a single user request, including all follow-up clarification rounds and parameter corrections. Generate the session-id once at the very first step and keep the same value throughout — even if the user provides new parameters, replies to a clarifying question, or the workflow branches. If the agent loses context, recover the original session-id from the conversation history (e.g., from a previously issued aliyun command's --user-agent value) rather than generating a new one.
Format: a 32-character lowercase hexadecimal string (for example, a1b2c3d4e5f60789a1b2c3d4e5f60789).
Reuse the same session-id consistently across every invocation method used in the session — whether the request is issued via the Alibaba Cloud CLI, an SDK, or Terraform.
Keep the session-id stable for the whole session, and use a fresh session-id for each new session rather than reusing one across sessions.
When an operation fails, report the session-id alongside the RequestId so the failure can be correlated between the User-Agent and the Alibaba Cloud API request log.
Handle errors and retries
[MUST] Environment Integrity. When executing CLI commands:
Leave environment variables that control CLI behavior (e.g., ALIBABA_CLOUD_CLI_MOCK, MOCK_MODE, CLI_PROFILE, ALIBABA_CLOUD_CREDENTIALS_FILE) untouched — the runtime environment is pre-configured; altering it breaks observability and testing hooks.
Use only documented flags; add flags such as --no-mock, --skip-verify, or --dry-run only when explicitly requested by the user.
Issue CLI invocations directly, rather than wrapping them inside shell scripts, subshells, or env -i launchers that strip environment context.
Every aliyun ecs <subcommand> call is issued as a direct, single-line command with all parameters inline — no indirection, no piping through intermediate scripts.
If an API call returns an unexpected success when an error was anticipated, report the anomaly and re-check the command format rather than silently proceeding.
Map common errors to concrete actions:
[MUST] Report failures before continuing. When any aliyun call returns a non-zero exit code or an error body, first execute the matching handling rule below and report the error code, RequestId, and affected resource IDs to the user before proceeding to the next workflow step or declaring the task successful. Surfacing every error keeps the user informed and the workflow trustworthy.
[MUST] Four-class error decision table — execute EXACTLY as written, no improvisation. Before handling any failed call, extract two fields verbatim from the error output: ErrorCode and RequestId. Then match the ErrorCode string against the four classes below in order and execute the prescribed steps literally. Preserve the RequestId and echo it verbatim in the final answer for every failure — keep it complete and unaltered rather than paraphrasing, truncating, or omitting it.
Class 1 — Throttling (identification: ErrorCode contains Throttling, e.g. Throttling.User):
Approach: retry with exponential backoff, up to 5 attempts — wait before each retry rather than retrying immediately, stay on the SAME command rather than switching to a different API, and give the retries a chance rather than giving up after the first hit.
Steps: retry the SAME command with exponential backoff — wait 1s before retry 1, 2s before retry 2, 4s before retry 3, 8s before retry 4, 16s before retry 5 (max 5 retries total). Use sleep 1 / sleep 2 / sleep 4 / sleep 8 / sleep 16 between attempts. If it still fails after 5 retries, stop, report ErrorCode + RequestId to the user, and ask whether to continue later. If a retry succeeds, continue the workflow normally.
Class 2 — Server-side internal error (identification: ErrorCode contains InternalError, ServiceUnavailable, or the HTTP status is 5xx):
Approach: retry up to 3 times rather than treating the first hit as fatal, keep the RequestId rather than discarding it, and surface the failed step rather than silently skipping it.
Steps: retry the SAME command at most 3 times with a fixed interval of about 2 seconds between attempts (sleep 2). Preserve the RequestId from the FIRST failure. If a retry succeeds, continue the workflow (e.g., keep polling to completion). If all 3 retries fail, stop, report ErrorCode + the preserved RequestId, and suggest opening an Alibaba Cloud ticket with that RequestId.
Class 3 — Parameter / client error (identification: ErrorCode contains ParameterInvalid, InvalidParameter, InvalidDiskId.NotFound, InvalidSnapshotId.NotFound, InvalidInstanceId.NotFound, or the HTTP status is 4xx and it is not a permission or throttling error):
Approach: validate first rather than blindly re-running the same failed command, keep the user's resource ID rather than substituting a different one on the assistant's own initiative, and resolve the error before moving to later steps or declaring the task complete.
Steps: first call the matching read-only validation API (DescribeDisks for disk parameters, DescribeInstances for instance parameters, DescribeSnapshots for snapshot parameters) in the same region to verify the resource exists and the parameter format is correct. If validation confirms the resource exists and the parameters are correct, retry the original command once. If validation shows the resource does not exist (TotalCount=0 or empty list), STOP execution, report the exact resource ID + RegionId + ErrorCode + RequestId to the user, and wait for the user to supply a valid ID — waiting for the user is the right path here, in preference to retrying or swapping in another resource.
Class 4 — Permission error (identification: ErrorCode contains Forbidden — e.g. Forbidden.RAM — or NoPermission, or the HTTP status is 403):
Approach: permission failures are NOT transient, so hold off on re-running the same API (in identical form or with changed parameters, casing, pagination, or action-name variant) until the user confirms authorization is complete; use aliyun configure list rather than reading credential files to diagnose; when a query cannot complete, report that fact rather than presenting fabricated results or declaring success. This no-retry approach applies only to the API that returned the permission error — other, unrelated APIs that the workflow legitimately needs can proceed.
Steps: stop immediately on the first permission error. Report to the user, all in the final answer: (1) the exact error code string (e.g. Forbidden.RAM), (2) the complete RequestId copied verbatim from the error body, (3) the affected operation and resource IDs. Then provide RAM authorization guidance: state that the current credentials lack the required permission (e.g. ecs:DescribeSnapshots), recommend attaching the managed policy AliyunECSReadOnlyAccess or a custom policy containing ecs:Describe*, and direct the user or their administrator to complete authorization in the RAM console before retrying. Wait for the user's confirmation before retrying.
Additional per-error rules:
IncorrectDiskStatus / IncorrectInstanceStatus: [MUST] immediately call DescribeDisks/DescribeInstances to fetch the current state, report it to the user, and then either wait for the blocking state to clear (e.g. an in-progress snapshot reaching accomplished) and retry, or ask the user how to proceed. Do NOT skip the step, do NOT fall through to later steps, and do NOT declare success while the error is unresolved.
SnapshotCreatedImage: block deletion and list the image IDs that must be removed first.
SnapshotCreatedDisk: require Force=true and explicit confirmation.
CreateSnapshotGroup failures — keep the consistency-group workflow rather than silently downgrading to a single-disk snapshot. When the request is for a consistency group, keep the consistency-group workflow and diagnose the specific error rather than falling back to CreateSnapshot:
InvalidInstanceId.NotFound (or 404): call DescribeInstances to confirm the instance exists and that its region matches the requested RegionId. Report the mismatch and ask the user to correct it.
DiskCategory.OperationNotSupported (or 400): call DescribeDisks to confirm all target disks are ESSD series and located in the same availability zone. Report exactly which disks fail the condition.
On any validation failure, report the specific unmet condition to the user and request correction. Keep CreateSnapshotGroup as the target action rather than silently replacing it with CreateSnapshot. Only after CreateSnapshotGroup succeeds, poll DescribeSnapshotGroups until Status=accomplished.
QuotaExceed.Snapshot / QuotaExceed.AppliedAutoSnapshotPolicyQuota: run cleanup planning or remove unused policies.
InvalidParameter.Mismatch during rollback: verify encryption and source-disk consistency.
InvalidAccountStatus.NotEnoughBalance: pause operations and advise account recharge.
InvalidDiskId.NotFound / ParameterInvalid on ApplyAutoSnapshotPolicy: follow Class 3 above — verify disk existence via DescribeDisks in the specified region; if the disk is missing (TotalCount=0), halt execution and inform the user with the exact DiskId and RegionId, leaving the task unmarked until a valid ID arrives.
404 Not Found on signed API POSTs while the endpoint host is reachable (a bare curl to the host returns 400/normal): this is usually a recoverable local environment issue, most commonly a proxy interfering with the signed request — not a skill, credential, or CLI-version bug. Before giving up, try these in order: (1) Clear proxy variables and retry by prefixing the command with HTTPS_PROXY= HTTP_PROXY= ALL_PROXY= no_proxy="*" — this resolves the majority of such 404s; (2) if it persists, verify DNS resolution and endpoint format (ecs.<region>.aliyuncs.com vs the global ecs.aliyuncs.com); (3) only after the above are ruled out, conclude the environment genuinely cannot reach the ECS OpenAPI, inform the user, and fall back to conceptual guidance. Do not blindly loop through many alternate endpoints, but do perform the proxy-clear retry — it is the single most common fix.
Server-side 5xx errors: follow Class 2 above — retry up to 3 times with ~2s interval, preserve RequestId, and suggest opening an Alibaba Cloud ticket.
When an operation fails, always include the RequestId in the response so the user can provide it to Alibaba Cloud support.
Response status: confirm a RequestId is present (indicates a successful API call).
Create: poll DescribeSnapshots / DescribeSnapshotGroups until Status is accomplished and Available is true.
Delete: re-query and confirm the snapshot no longer appears in the inventory.
Rollback: poll DescribeDisks (NOT DescribeSnapshots) with the target DiskId until the disk status returns to In_use/Available, and, if applicable, confirm the instance starts successfully.
Auto policy: query DescribeAutoSnapshotPolicyEx and confirm Status is Normal and DiskNums reflects the applied disks.
No blocking warnings: review any warnings and unmet preconditions before reporting success.
Cleanup
Snapshot creation, snapshot preheating, snapshot replication, and auto-snapshot policies may incur storage, preheating, cross-region traffic, or policy charges — remind the user to delete unneeded snapshots, disable unused cross-region copy, and remove unused policies to control cost.
Read-only operations (list/inspect/cost) create no resources and require no cleanup.
Best Practices
Create snapshots during low-traffic windows to minimize I/O impact.
Always set meaningful SnapshotName, Description, and tags for traceability.
Use RetentionDays to avoid runaway storage costs.
Poll until Status is accomplished before relying on a snapshot for rollback or image creation.
Always create a pre-rollback backup snapshot before performing a rollback.
Use dry-run validation to check rollback preconditions before the actual call.
The following scripts are provided as local reference tools only. They do not execute destructive operations, and the assistant does not run them on the user's behalf. They may be referenced when discussing non-destructive tasks such as cost estimation or parameter validation.
Validate policy parameters and generate sample values.
Common Issues and Solutions
Issue: "IncorrectInstanceStatus" when rolling back a disk.
Solution: Stop the instance first; ResetDisk/ResetDisks require the instance to be Stopped.
Issue: "SnapshotCreatedDisk" when deleting a snapshot.
Solution: A disk was created from this snapshot. Confirm with the user, then set Force=true. The created disk can no longer be reinitialized afterward.
Issue: "SnapshotCreatedImage" when deleting a snapshot.
Solution: The snapshot backs a custom image. Delete the image first, then delete the snapshot.
Issue: Every API call returns 404 Not Found, but the ECS endpoint host is reachable (bare curl returns 400).
Solution: This is usually a local proxy interfering with the signed request, which is recoverable. First retry with proxy variables cleared: HTTPS_PROXY= HTTP_PROXY= ALL_PROXY= no_proxy="*" aliyun ecs <Action> ... — this fixes most cases. If it still fails, check DNS resolution and endpoint format. Only after proxy and DNS are ruled out should you conclude the environment cannot reach the ECS OpenAPI and fall back to conceptual guidance. Do not treat the first 404 as unrecoverable.
Issue: CreateSnapshotGroup fails.
Solution: Confirm all disks are ESSD series, in the same availability zone, not multi-attach, and within the 16-disk group limit.
Issue: "QuotaExceed.Snapshot".
Solution: Identify deletable snapshots conceptually, or request a quota increase.
Issue: Permission / "Forbidden" error.
Solution: Follow the Permission Failure Handling process — read references/ram-policies.md and use the ram-permission-diagnose skill.
