The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill identifies and directs users to flutterreleases.com as a primary source for Flutter release tracking and archive manifests. This specific domain has been flagged as malicious and blacklisted by automated security scanners.
[METADATA_POISONING]: The skill presents fictional or future-dated framework milestones (e.g., Flutter 3.50 in 2026) with an authoritative tone. It incorrectly claims that core SDK libraries like package:flutter/material.dart are deprecated, directing users to migrate to external packages such as material_ui and cupertino_ui. These claims are not supported by current official Flutter documentation and could lead users to install unnecessary third-party dependencies under the guise of an official update.
[INDIRECT_PROMPT_INJECTION]: The skill defines runbooks that process project configuration files (pubspec.yaml) and source code, creating a surface for indirect prompt injection. An attacker-controlled project could use malicious code structures to influence the agent's behavior during migration tasks.
Ingestion points: Project pubspec.yaml, toolchain version outputs, and Dart source code.
Boundary markers: Absent; there are no instructions to ignore embedded commands in the processed content.
Capability inventory: The skill's instructions involve the execution of shell commands such as dart fix, flutter pub add, and platform-specific build commands like pod deintegrate.
Sanitization: No validation or sanitization of external project data is performed before it is used to guide the migration process.

flutter-sdk-changelog