tessri.
  • Zero dependencies
  • Offline, reads git
  • Never edits your code

Keep code understandable.

Agents write more of the code; people still have to stand behind it. Tessri reads a codebase and its git history, offline and with nothing else installed, and tells the person a change is handed to where to look first, who is able to look, what it did to the structure, and what nobody maintains, with the evidence for every claim.

$ pip install tessri

Python 3.11 or newer, no dependencies. To run it once without installing: uvx tessri scan .

$ tessri brief --since 12e6c2cd6

12e6c2cd6 (base 12e6c2cd6) to HEAD fbb6833c5. 4 code files changed (0 added, 4 modified, 0 deleted), 429 lines; 10 other files (249 lines of config, docs and the like).

Look first

Ranked by history: commits in 90 days, fixes and reverts, and who went quiet. The owners are the people who changed the file, up to the base.

#FileWhyThis changeOwnersFlags
1tools/build_site.py15 commits in 90 days+251 -12Tianyi Zhang 100%-
2site/index.html15 commits in 90 days+5 -5Tianyi Zhang 100%-
3site/tessri.css9 commits in 90 days+83 -1Tianyi Zhang 100%-
4tests/test_site.py6 commits in 90 days+71 -1Tianyi Zhang 100%test file

Nobody other than the authors maintains the changed files, as far as history shows.

What it did to the structure

What the change did to the import graph, from one analysis of the base and one of the working tree: orphans, cycles, dangling imports, changed code nobody active maintains, and what outside the change imports it.

No structural findings: the change adds no orphan, cycle, dangling import or unowned code that this analysis can see.

Nothing outside the change imports it, as far as static imports show.

Not analysed: site/tessri.css (CSS, no analyzer), so dependents, candidate tests and structure are unknown for it.

Not analysed: site/index.html (HTML, no analyzer), so dependents, candidate tests and structure are unknown for it.

Validation

The candidate tests are paths that reach the change; a check counts as run only when the host recorded it. Nothing here claims that a test passed.

CandidatesCandidate tests, nearest first: tests/test_site.py. They reach the changed files; they are paths, not results.

RanChecks the host ran: none recorded. Record each with --ran "<command>: <result, exit code>", or hand over a test report with --report or a coverage report with --coverage.

The latest change on Tessri's own main is pull request #29, “The front doors do more: the site searches the guide, pages know their neighbours and their card, and the repository explains itself”, merged as fbb6833c5. Three of the page's five sections; the whole page adds the commits, what needs a person and the caveats, and the HTML report has the module graph and the impact explorer.

Open the full demo
What it answers

Four questions, answered from one picture of the code and its history, and one page that hands a change over.

One page that hands a change over.

Where to look first, who is able to look, what the change did to the structure, which tests reach it and which checks ran, and what needs a person: from one analysis of the base and one of the working tree, with the evidence for every line.

  1. Look firstThe changed files, ranked by history: commits in 90 days, fixes and reverts, and who went quiet.
  2. What it did to the structureNew orphans, cycles, dangling imports, changed code nobody active maintains, and what outside the change imports it.
  3. ValidationThe tests that reach the change, kept apart from the checks the host recorded as run.
  4. Open for a personWhat the analysis cannot settle: a decision, a second reader, a check nobody ran.
  5. Read this with careWhat every line rests on, and what it cannot see.
tessri brief --since main

How does the system fit together?

The modules, the cycles, and the few dependencies that would break them if cut.

tessri graph

What can this change reach?

The files that depend on it, the tests that reach it, and who is able to review it.

tessri impact <path>

What does nothing reach, and what does nobody maintain?

Every file's status from entry points, tests and configs; owners from history, and who went quiet.

tessri scan .

What did this change do to the structure?

New orphans, new cycles, imports of deleted files, changed code nobody active maintains. Markdown, JSON or SARIF, with an exit code for CI.

tessri check
A complete example

What can a change to one file affect?

Real output on Tessri's own repository, as the Markdown renders.

$ tessri impact tessri/graph.py

Changing tessri/graph.py (1 file, 113 code lines) can affect:

  • 19 code files in 2 modules (46.3% of the code), 1 of them directly (one followed edge away)
  • 25 test files reaching it (candidates to run, not runs), 1 directly
  • code files by number of followed edges away: 1: 1, 2: 14, 3: 4
  • edges followed, any number of steps: recognized dependency syntax, package files, and 3 references found in strings, dynamic loading or unclassified evidence (for example tests/test_site.py → tessri/cli.py, string); syntax and package files alone reach 19 code files and 25 test files
Affected moduleFilesDirectOwnersState
tessri161Tianyi Zhang 100%owned
tools30Tianyi Zhang 100%owned

People who maintain the affected code (tell them, or ask them to review): Tianyi Zhang 100%

Tests that reach it, nearest first: tests/test_site.py, tests/test_htmlreport.py, tests/test_javascript.py, tests/test_markup.py, tests/test_meta.py, tests/test_modules.py, tests/test_moves.py, tests/test_ownership.py, tests/test_provenance.py, tests/test_renames.py and 15 more

It depends on: tessri (2), tessri/analyzers (1)

Static analysis of this repository under the selected edge policy and hop limit: registries, dependency injection and consumers outside the repository are not counted, and a file outside this set is not shown to be unaffected.

The reach of tessri/graph.py, drawn from this build's analysis, each edge in the kind of evidence behind it.

an import statementa name in a string, a config or a dynamic import: a guess

the file imports it 1 file one step further 14 files graph.py analyze.py brief.py cli.py explain.py hook.py htmlreport.py mcp.py meta.py modules.py report.py +5 more

Candidates27 test files reach it, paths to run, not runs: test_site.py, test_cache.py, test_htmlreport.py, test_javascript.py, test_markup.py, and 22 more.

That's it. tessri why tessri/graph.py shows the evidence behind any line of it, and --edges imports keeps to the recognized imports and leaves the string references at the frontier.

How does it work?

One pass over the files and the history, then one view per question.

No model, no network, no service, and it never edits your code. The order triage puts files in was backtested on real history; the composite risk score we tried did not beat its parts, so there is none.

01

Read

Lists the files, parses Python with Python's own parser and JavaScript and TypeScript with its own scanner, and reads configs, scripts and docs for the files they name.

02

Join

Builds the import graph with the kind of evidence behind every edge, and reads from git who changed what, following files through renames.

03

Answer

Each command is one view of that picture: reachability, ownership, module graph, impact, triage and the check of a change.

Evidence, not adjectives

Every claim can be traced, and what the analysis cannot see it says it cannot see.

17 test files reaching it (candidates to run, not runs)tessri impact
13 names in configs, scripts or code fit several files each and so name none of them: utils.py (15), config.py (12), …tessri scan, caveats
Dependencies are static imports between files of this repository. A deleted or moved file that an untouched config, script or document still names, or that untouched code still names in a string or a dynamic import, is reported with the kind of each name; a name in a string is a guess that the file is loaded at run time, so that one is for a person. Registries and consumers outside the repository are invisible, so a clean result does not mean nothing else depends on the change.tessri check
Getting started

Three steps.

  1. Scan a repository. tessri scan . prints the report; --format html --out report.html writes one offline page with the explorer.
  2. Ask before, during and after a change. tessri impact <path> before, tessri triage to review one, tessri check when it is done, tessri brief to hand it over on one page; without --since, a branch is judged against the default branch.
  3. Give it to your agent, and to every pull request. tessri integration install --agent claude (or codex) installs the skill with a receipt, --agent review posts the brief on every pull request; tessri integration remove takes it all out again.