PhilipThomas6w/claude-tools

Build tooling for AI features: build-agent, build-connector, build-mcp-server and write-evals skills, plus a manifest-checker agent. Composes with loop-harness.

Discovery and design process for AI projects specifically, not a general project process. Requirements gathering captured as working markdown, with the document package rendered at each gate by the project's chosen style pack (default generic-docx). Includes /new-ai-project, /render-gate-package, the ai-poc-docs skill, discovery and gate-renderer agents, and a design-reviewer agent. Hands off to loop-harness for implementation.

Neutral, unbranded .docx style pack (A4, Arial, neutral palette). The default style pack for ai-project.

Stack-agnostic loop-engineering harness: pwsh hooks (session context, Bash/PowerShell deny-list guard, Stop+SubagentStop gate on build/verify.ps1 with a real round-trip cap and ledger logging), maker/checker/explorer/verifier sub-agents with real effort routing, /init-harness bootstrap, /work-next automation, per-stack verify templates that fail loudly until wired, and harness-audit/claude-code-hook-facts/verify-gate-authoring skills.