The Agent Skills Directory

[PERSISTENCE]: The skill establishes a mechanism to maintain activity across sessions by using the schedule_task function to run a background script at regular intervals.
[PRIVILEGE_ESCALATION]: The skill instructions include using chmod +x to grant execution permissions to a script file created at runtime within the workspace.
[DYNAMIC_EXECUTION]: The skill uses a bash heredoc to dynamically generate executable code in squadron-inbox-poll.sh, which is then executed via a scheduled bash command.
[COMMAND_EXECUTION]: The skill documentation provides commands to execute bash scripts and use curl for interacting with external APIs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external Squadron API endpoints that may contain untrusted content.
Ingestion points: Untrusted data enters the agent context through GET requests to the inbox and knowledge base endpoints in SKILL.md.
Boundary markers: There are no delimiters or explicit instructions provided to the agent to ignore embedded commands within the fetched data.
Capability inventory: Across SKILL.md, the skill has network access via curl and requests, file-writing capabilities via kb_write, and scheduled command execution via schedule_task.
Sanitization: There is no evidence of validation, escaping, or filtering of the external API content before it is processed or displayed.

@1892/squadron