The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill directs the agent to locate and install missing specialist skills from the external domain skills.sh using npx. This allows for the runtime download and execution of arbitrary code from a third-party source that is not recognized as a trusted provider.
[EXTERNAL_DOWNLOADS]: The workflow depends on several external automation agent skills, including stablyai/orca@computer-use, midscene-skills@computer-automation, and codex-skills@gemini-computer-use. These components originate from unverified external repositories.
[COMMAND_EXECUTION]: The skill facilitates the execution of shell-based commands through npx and delegates operations to sub-agents capable of system-level automation.
[PROMPT_INJECTION]: The skill acts as an orchestrator for untrusted data, creating an indirect prompt injection surface. 1. Ingestion points: The skill reads from .agents/startup_dna.md and processes direct user requests. 2. Boundary markers: The skill mentions enforcing security gates for releases but lacks explicit input delimiters or instructions to ignore embedded commands in the data it processes. 3. Capability inventory: The skill can invoke sub-agents and execute remote code installation commands. 4. Sanitization: There is no evidence of validation or sanitization of input data before it influences orchestration logic or script installation.

startup-workflow-orchestrator