The Agent Skills Directory

[Command Execution Capability]: The skill utilizes local shell commands including git, make, and uv to manage code repositories, install dependencies, and deploy application services. These commands are integral to the application lifecycle management tasks the skill performs.
[External Resource Integration]: The skill references and downloads content from official GitHub repositories and documentation sites associated with the underlying SDK. These resources provide necessary configuration and examples for the deployment manager and SDK usage.
[Indirect Prompt Injection Surface]: The skill is designed to ingest data from external sources such as prior session links, summaries, and JSONL rollout files to generate application code. This introduces a surface where instructions embedded in that data could influence the agent's behavior.
Ingestion points: Processes Codex threads, session links, rollout JSONL paths, and pasted summaries (SKILL.md).
Boundary markers: Instructions specify creating a compact brief for verification (SKILL.md).
Capability inventory: Performs file writes, shell executions via uv run and make, and network operations via git and curl (SKILL.md).
Sanitization: Requires explicit user approval before implementing code based on a brief (SKILL.md).
[Dynamic Code Execution]: To facilitate testing and evaluation, the skill executes generated code locally within the development workspace. This allows for runtime verification of agent logic and tool interactions before deployment.
[Integrated Secret Management]: The skill includes a dedicated flow to interface with a platform-specific API key skill, ensuring that sensitive credentials are not hardcoded or exposed in logs during the development process.

agents-sdk