The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is authorized to run common development and build tools via the Bash tool, including git, make, npm, cargo, and dotnet. This is necessary for its primary function of code exploration, testing, and implementation within a repository environment.
[EXTERNAL_DOWNLOADS]: The use of package managers such as npm, pnpm, yarn, go, and cargo typically involves downloading dependencies from well-known public registries during the build or test phases.
[PROMPT_INJECTION]: The skill processes external feature briefs which could potentially contain malicious instructions (Indirect Prompt Injection).
Ingestion points: Untrusted data enters the context via the $ARGUMENTS variable or the contents of a local .md brief file.
Boundary markers: No explicit delimiter or "ignore instructions" warning is used for the brief content.
Capability inventory: The skill has access to file writing (Edit, Write), shell execution (Bash), and subagent creation (Agent).
Sanitization: There is no technical sanitization, but the risk is significantly mitigated by the mandatory human checkpoint in Phase 4, where the user must approve the generated plan before any execution occurs.
[DATA_EXFILTRATION]: The skill includes specific instructions to the agent to avoid staging or committing sensitive files like .env or credentials, and to avoid echoing secrets into the shell, which follows security best practices.

plan-and-build