The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to install or update the Alibaba Cloud CLI using a script fetched from the vendor's official CDN at https://aliyuncli.alicdn.com/setup.sh. As this resource is owned by the vendor (aliyun) and is required for the skill's operation, it is considered a legitimate administrative action.
[COMMAND_EXECUTION]: The skill involves the execution of numerous Bash commands through the aliyun CLI to perform diagnostics and repairs. This is the primary mechanism by which the skill functions.
[PRIVILEGE_ESCALATION]: The skill has the capability to modify, create, and attach Alibaba Cloud RAM policies to users and roles. This functionality is the core purpose of the skill—allowing the agent to resolve 'Access Denied' errors by escalating or correcting permissions under user guidance.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses error messages and encoded diagnostic strings from external CLI outputs. While this represents a surface for indirect prompt injection if the output were tampered with, the skill uses structured diagnostic logic to interpret the data.
Ingestion points: Extracts error_code, missing_action, and encoded_message from raw CLI error responses in SKILL.md and references/diagnose-flow.md.
Boundary markers: None explicitly used to wrap ingested error data.
Capability inventory: Administrative access to RAM identities and policies (create-policy, attach-policy-to-user, update-role) via aliyun CLI as defined in references/ram-cli-commands.md.
Sanitization: The skill relies on transcription and parsing of specific fields without explicit sanitization steps for the ingested strings.

alibabacloud-ram-permission-diagnose