Skip to main content

vulkro ctx

A code-reading server for your repository, for any AI coding agent: Codex, Claude Code, Cursor, GitHub Copilot, Gemini CLI, Aider and any other agent that can run a shell command. Each answer is just the code that answers the question, as numbered text an edit can target.

vulkro ctx read FILE            # a small file whole; a large one as an outline of its declarations
vulkro ctx read FILE:NAME       # one function, class or type (FILE:Type.method works)
vulkro ctx read FILE:120-180    # a line range
vulkro ctx search NAME          # every definition, call and use of NAME, in one answer
vulkro ctx search NAME --lines  # the same, locations only
vulkro ctx body NAME            # one function and the lines that call it
vulkro ctx impact NAME          # everything that depends on NAME, and the tests to run
vulkro ctx map                  # the most central functions, to get oriented
vulkro ctx check                # after an edit: what it broke
vulkro ctx init                 # tell your repository's agents to use these

read works on one file in every supported language, Rust included. search, body, impact and map use the call graph, kept warm by a small background server per repository, so an answer takes a few tens of milliseconds. The server listens only on a local socket, makes no network calls, and stops after 15 idle minutes.

Tell your agents​

vulkro ctx init

adds a short block to AGENTS.md, the instruction file most agents read, and to CLAUDE.md, GEMINI.md or .github/copilot-instructions.md if your repository has them (--agents cursor writes a Cursor rule). Running it again updates the block in place.

What did my change break?​

vulkro ctx check reads git diff and reports, in one answer:

  • call sites that still pass the old arguments to a function whose parameters changed, with the enclosing function and the line;
  • edited calls aimed at a same-named method on another type;
  • removed functions and types that are still used;
  • syntax errors on the changed lines, such as a bulk edit that mangled a call;
  • calls whose positional arguments now land on a different parameter, for review;
  • what depends on the change, and the tests to run;
  • security findings on the lines you changed (--fast skips the scan).

It exits 1 when it finds a problem. vulkro install-hook --pre-commit runs it on every commit (see install-hook), so it catches what an agent missed whichever model made the change. MCP clients get the same answers from the read_code and check_change tools.

Measured​

  • On 13 everyday tasks across four repositories, agents given only the vulkro ctx commands took 49% fewer turns and read 37% less code than the same agent searching and reading files itself, at about 15% lower cost, with the same results (vulkro 0.33.0, October 2026).
  • Over 62 recorded agent edits of real Go code, replayed and compiled, vulkro ctx check flagged all 10 that left the build broken and none of the 52 that did not.

The benchmark, its tasks and its scoring are in the product repository under bench/agent_context.

Last updated on