Investigate Gitingest integration for compact LLM project context

[norberia]

Summary

Investigate integrating Gitingest to turn a repository or local project into a compact, prompt-friendly text digest for LLM analysis.

Gitingest supports local directories and repository URLs, exposes a CLI and Python package, and produces repository summary, directory structure, file contents, and token-count information. The investigation should determine whether it improves context efficiency and analysis quality in the skill workflow.

Research scope

• Compare Gitingest output and token usage with the current project discovery and file-reading approach.
• Evaluate local CLI/package integration versus use of the hosted service; project contents must not be sent to an external service by default.
• Define safe include/exclude behavior for generated files, dependencies, binaries, ignored files, credentials, secrets, and other sensitive content.
• Test large repositories, monorepos, unsupported/binary files, submodules, and projects near the LLM context limit.
• Determine how token budgets, truncation, prioritization, and incremental refresh should work.
• Identify the best integration phase and whether the digest should be ephemeral or stored as a generated artifact.
• Assess installation requirements, version pinning, license, maintenance health, performance, and failure fallback.
• Prototype the integration on representative projects and measure token reduction, runtime, and retained analysis quality.

Deliverables / acceptance criteria

• A documented security and architecture recommendation for local versus hosted execution.
• Reproducible benchmark results covering token count, latency, and useful-context retention.
• A proof of concept or a clear rejection rationale.
• A proposed integration contract, including inputs, outputs, exclusions, size limits, errors, and fallback behavior.
• An implementation and eval/test plan if adoption is recommended.

References

• Gitingest website
• Gitingest source repository