The Agent Skills Directory

[PROMPT_INJECTION]: The static analysis detected phrases such as 'ignore previous instructions' in references/failure-modes.md. Upon manual review, these are identified as educational examples within a 'Trust boundary' advisory section. The text warns users about the potential for malicious Slack messages to influence agent behavior (indirect prompt injection) and provides remediation guidance, rather than attempting to inject instructions into the agent itself.
[SAFE]: The skill is strictly read-only and relies on the official Slack MCP server for all operations. No hardcoded credentials, suspicious network calls, or remote code execution patterns were found. The use of Markdown for message rendering is consistent with expected agent behavior, and the author proactively documentation threat models related to delegated authorization and workspace visibility.

slack-automate