The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill has a high attack surface for indirect prompt injection by processing external data from a Slack workspace.
Ingestion points: The scripts channel_history.py, search_messages.py, and search_files.py retrieve untrusted strings from Slack.
Boundary markers: Absent; there are no instructions to the agent to treat Slack content as data rather than instructions.
Capability inventory: The agent can execute send_message.py, delete_message.py, and upload_file.py, which could be exploited if the agent follows instructions found in Slack history.
Sanitization: Absent in the provided markdown workflow.
[Command Execution] (SAFE): The skill executes local scripts to perform its tasks.
Evidence: Commands like python 00-system/skills/slack/slack-master/scripts/send_message.py are used.
Context: These are internal skill scripts and do not appear to be dynamically generated from untrusted input in a dangerous manner.

slack-connect