The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands bun run tsc --noEmit and bun test to verify code integrity during the development process.
[COMMAND_EXECUTION]: It utilizes git for staging changes and creating incremental commits that pair code modifications with specification updates.
[SAFE]: Implementation tasks are isolated using sub-agents, which helps ensure that each coding task has a restricted context and does not affect unrelated parts of the codebase.
[SAFE]: The process requires the agent to present an implementation plan to the user for approval before beginning any execution waves.
[SAFE]: The skill ingests external data from specification files (specs/*.md), creating a surface for potential indirect prompt injection. This risk is minimized by the modular nature of the sub-agent prompts and the explicit manual approval steps in the workflow. Ingestion points: specs/*.md files; Boundary markers: None explicitly defined; Capability inventory: File system read/write, sub-agent creation, and execution of local shell commands (git, bun); Sanitization: No explicit sanitization of spec content is described.

spec-execution