slack

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCECOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external Slack messages and direct messages, which are ingested via the slack_listen.py daemon and slack_read.py script. This content is interpolated into the agent's prompt context. Although soul_engine.py implements boundary markers using XML tags and markdown code blocks to fence user input, the agent's high-privilege capabilities—including shell command execution (Bash) and file system access (Read/Write) as listed in skills.md—create a surface for indirect prompt injection attacks.
  • [PERSISTENCE]: The skill includes a macOS LaunchAgent configuration (com.minoan.slack-daemon.plist) and associated installation scripts in the daemon/launchd directory. These are intended to maintain the Slack bot as a persistent background process across system restarts.
  • [COMMAND_EXECUTION]: The daemon/claude_handler.py script executes the Claude CLI via subprocess.run. Additionally, the skill explicitly enables the Bash tool for the agent, providing the capability to execute shell commands within the host environment.
  • [DYNAMIC_EXECUTION]: Multiple scripts, such as scripts/slack_channels.py, scripts/slack_post.py, and scripts/slack_users.py, utilize dynamic imports via __import__("os") to perform runtime path resolution for resolving local module locations.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 06:34 PM