The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to automatically ingest and act upon data (task titles and descriptions) fetched from an external source (Linear API). This creates a surface for indirect prompt injection where malicious instructions embedded in a project management task could influence the behavior of the coding agent.
Ingestion points: Symphony.Linear.Client.fetch_todo_tasks in SKILL.md fetches issues from the Linear GraphQL API.
Boundary markers: The provided Elixir code snippets do not show any delimiters or 'ignore embedded instructions' warnings when passing task data to the Codex.implement function.
Capability inventory: The agent has the capability to create git branches, modify the codebase (via Codex), and open Pull Requests on GitHub.
Sanitization: No sanitization or validation of the task description is performed before it is processed by the implementation agent.
[EXTERNAL_DOWNLOADS]: The skill configuration and reference implementation involve cloning code and fetching specifications from external repositories.
Evidence: The setup instructions include git clone https://github.com/openai/symphony.git and references to SPEC.md hosted on GitHub.
Note: These references target an official repository for the tool's implementation and are part of the standard installation workflow.

openai-symphony-autonomous-agents