ce-slack-research

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The prompt explicitly instructs agents to ignore/enforce that "tool/model limits are NOT otherwise enforced once de-plugin-ified" (an attempt to override system/tool-level constraints) which is outside the Slack-research skill's stated purpose, so this is a prompt-injection-like override; no other hidden/obfuscated instructions are present.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.95). Outsider-authored free text from Slack messages is ingested at runtime via the Slack MCP tools (slack_search_public_and_private → slack_read_thread / slack_read_channel), and then summarized/synthesized into the agent’s LLM context.