The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from untrusted external sources, such as public GitHub repositories, URLs, and issue trackers. This creates a surface where malicious instructions embedded in the external content could potentially influence the agent's behavior during the plan generation or company creation phases.
Ingestion points: External Git repositories, tweets, and GitHub issues.
Boundary markers: The skill contains instructions to treat external data as hostile and redact secrets, though it lacks strict data-format delimiters in its prompt templates.
Capability inventory: The skill has access to file system writing (via MemoryStore), network operations (via git and gh tools), and shell command execution.
Sanitization: There is an explicit instruction for the agent to redact PII and secrets before sharing context with subagents.
[DYNAMIC_EXECUTION]: The implementation workflow (in the writing-plans and executing-plans sub-skills) involves the automated generation of implementation steps that include shell scripts and application code, which are then executed by the agent. While these are generated by the AI agent, they are derived from potentially untrusted Product Requirement Documents (PRDs) or repository analyses.
[COMMAND_EXECUTION]: The skill uses various command-line tools for its operations, including git for repository cloning and metadata retrieval, gh for issue management, and language-specific test runners like pytest and npm. Execution is performed through a specific proxy interface (ExecutionProxy) using the rtk prefix to manage overhead and execution environment.

product