PROJECT DOSSIER
Cross-Platform Agent Engineering Harness (Claude Code / Codex)
The problem
A personal engineering harness for Claude Code and OpenAI Codex, sharing rules, skills and tools while adapting execution entry points and checks to each platform.
Engineering approach
- Derive platform entry points from shared rules, skills and tools; adapt tool interfaces and session lifecycles.
- Integrate structured outputs, tool calls and quality checks into development, document and data tasks, using multi-agent collaboration where appropriate.
- Maintain smoke, golden-task and reporting workflows; record platform differences and actual execution evidence separately.
Current results
- Shared skills and engineering procedures are reused across Claude Code and Codex for everyday development and delivery.
- Implemented Codex entry-point generation, hook protocol adaptation and native session integration; validation is path-specific, not a claim of full feature parity.
What cross-platform support means
Rules, skills and shared tools have one maintained source. Claude Code and Codex use platform-specific entry points and protocol adapters. Session history stays with each platform's native interface; hook coverage is verified through execution evidence.
This is personal engineering infrastructure used in real development and document workflows. It is not presented as a commercial deployment for external customers. Useful evidence includes a task's input, the tool calls, the checks that ran, and the resulting artifact; component counts alone do not measure effectiveness.
Tech stack
Claude Code · OpenAI Codex · MCP · Python · Shell · JSON Schema