The Agent Skills Directory

[SAFE]: The skill's primary function is to reduce the operational footprint and attack surface of the AI agent by disabling specific connectors, workflows, and tools via configuration.
[SAFE]: Security best practices are followed by using the AskUserQuestion tool to prompt for user approval whenever a target setting conflicts with an existing value in the local configuration file.
[SAFE]: The skill restricts its file operations to the local project-specific configuration directory (.claude/settings.local.json), avoiding global system impact.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface by reading the .claude/settings.local.json file. While this exposes the agent to potential indirect prompt injection, the risk is negligible as the skill processes the content as a structured JSON schema rather than executable instructions.
Ingestion points: Reads .claude/settings.local.json in step 1.
Boundary markers: None explicitly used, but logic relies on key-value pair validation.
Capability inventory: Uses Read, Write, Edit, and AskUserQuestion tools.
Sanitization: Employs mandatory conflict detection and user confirmation before applying changes.

claude-light