The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from web search results during the research phase (Phase 1), creating a surface for indirect prompt injection. This external data influences the design, planning, and code generation phases. 1. Ingestion points: Web search results for existing tools and API documentation (Phase 1, Step 2). 2. Boundary markers: Absent; the instructions do not specify the use of delimiters or warnings to ignore instructions embedded in retrieved data. 3. Capability inventory: The agent performs file system operations (creating directories, writing plans/code) and executes shell commands (npm/go initialization, linting) across all phases. 4. Sanitization: No explicit sanitization or validation of search content is mentioned before the data informs the agent's actions. Mitigation: The risk is significantly reduced by mandatory user approval checkpoints at the end of the research, design, and planning phases.
[COMMAND_EXECUTION]: The skill facilitates the execution of project initialization and validation commands, such as go mod init, npm init, eslint, and go vet. While these are standard development tools, they represent executable capabilities within the skill's workflow.
[EXTERNAL_DOWNLOADS]: The workflow incorporates the identification and recommendation of third-party libraries and dependencies based on web research and user requirements.

dev-new-tool