dsh-repo-analyzer

When you ask an agent to modify code in an unfamiliar repository, it typically burns a large number of tokens probing files one by one to figure out the tech stack, module boundaries, and dependency flow. dsh-repo-analyzer collapses that into a single millisecond-range local scan: it detects the tech stack via manifest files, parses dependency lists from package.json, pyproject.toml, go.mod, and Cargo.toml, and builds an import/require reference graph using heuristic regex matching — all in one structured JSON response. The agent can then identify architectural hot spots, assess change impact, and pinpoint which file to read first, without trial and error.

The design philosophy is zero extra services: no background processes, no LLM calls, no child agent spawns. Everything runs on the local filesystem via node:fs, with dsh-tools and schemastery as the only runtime dependencies. Security defaults are baked in — every user-supplied path is validated by resolveWithin to ensure it stays inside the configured root, node_modules, .git, dist, and build artifacts are excluded by default, oversized files are skipped automatically, and a hard maxFiles cap prevents runaway scans.

It fits any workflow where an agent needs to understand a repository before touching code: quickly mapping architecture when onboarding to a legacy project, evaluating how many modules reference a given dependency before a refactor, or letting the model independently locate the right starting file without wasting context window. The three tools — repo_scan, repo_deps, and repo_refs — each handle a distinct slice, and together they cover the full analysis chain from a high-level overview down to single-dependency impact.