Tessri guide
Tessri reads a codebase and its git history and answers the questions a maintainer asks of a system that grows faster than anyone can read it. One page per question, each with real output and with what the analysis cannot see.
- Getting started: install, run, or build a single-file copy.
- Which command, when: the moment, the command, the MCP tool, and the first thing to read in the output.
- What is unused: scan and reachability: every file's status, and how reachability is decided.
- Who maintains it: owners from git history, who went quiet, and
CODEOWNERSagainst reality. - Review load: how much review the recent changes ask for, where, and from whom.
- How the modules fit together: the module graph, the cycles, and the few dependencies to cut.
- What a change can reach: dependents, the tests that reach a file, the evidence behind each edge.
- Triage a change: which changed files deserve a person first, and who can look.
- Check what a change did to the structure: findings, the exit code for CI, and SARIF.
- Hand a change over: the review brief: triage and check on one page, from one analysis, with the state the evidence describes and what the host ran kept apart from candidate tests.
- The HTML report: one offline page with the explorer.
- Integration: skills, AGENTS.md, MCP, hooks, configuration, CI: every profile installed with a receipt, and taken out again.
- Hooks: Tessri while the agent works: what the agent hears after an edit and what you hear when it stops, once each, and how the hosts guard it.
- MCP: Tessri as tools:
tessri mcp, the eight tools, their arguments and errors, and the profiles that point each agent at the server. - Configuration:
.tessri.tomland[tool.tessri]. - JSON output: the objects,
schema_versionand themetablock. - Architecture: the pipeline, one view per command, and how to add a language.
- The backtest: how the order
triageuses and the one-import review set were tested, the script that runs the test on any repository, and its output on one. - The reviewer match: how who can look was tested against who actually reviewed, on a repository's own pull requests, and what that says about the list.
- Limits: what is static, what is invisible, and what to read with care.
- Formerly repolife: what to rename when upgrading from the old name.
Design contracts and acceptance records are in docs/design; copyable CI and agent recipes are in workflow-recipes.md; the examples of a review brief are in docs/examples.
Source: docs/guide/README.md, Markdown in the repository; this page is built from it. Edit it on GitHub.