Your repository is not ready for an agent to work in it.
And you don't know it, because nobody measures it. An agent that can't find where a responsibility lives, that can't run the tests, or that believes a stale doc isn't slow: it's dangerous.
No card. Your code never leaves your machine.
Application · PHP · 4,000 files · 520k lines
- Documentation38
- Agent instructions0
- Executability59
- History59
- Structure46
Open findings, with their cost in points
- −13.5Documentation coverage
- −6Agent instructions
- −5Concentrated knowledge
Real repositories in different languages, measured with the public rubric
The cost isn't the model. It's what your repository makes it read.
Every session starts by rebuilding the context the repository never gave it: tokens, minutes and wrong changes, every time.
Docs that lie cost double: the wrong change, and then fixing it.
With no tests that assert anything, the only verification available is your time.
Here's the state of the world, measured
- have no agent instructions at all
- 49%
- hides files that do not fit in a context window
- 88%
- does not cover half of what an agent needs to know
- 57%
More than 100 public repositories, measured with the same rubric you'd see on yours. The median scores 67 out of 100.
Same agent, same code, a fraction of the work
The same questions on the same commit, with and without the project documentation.
- fewer files to open before answering
- 81%
- less time to the answer
- 73%
Measured with real agents on the same commit: only documentation and agent adapters are removed, the code does not change.
And every finding becomes a card.
Your agent runs it and you approve. When the re-analysis confirms the fix, the card closes on its own.
How the loop worksDeclare the environment variables
−4 points
Split the 800-line files
Passes
Documentation coverage
feat/doc-coveragePassesYour code doesn't move. The one working in it is your agent, on your machine.
You connect the MCP client you already use and the deep analysis runs there, with your credentials. We send the questions; what comes back are observations citing file and line.
Your machine
Your repository, as it is
Your agent, with your credentials
The rubric's questions
Observations citing file and line
Your code
RepoFit
The rubric and the probes
Your score and your findings
- 1Connect the repository
- 2Get the score and the findings
- 3Your agent runs the remediation
- 4The score closes the card
We store git metadata, what your team writes and your agent's results. Never the contents of your files.
Read the full policyStart by knowing where you stand.
Connect a repository and get its score in minutes, with the findings and what each one is worth closing. Free, and no card.
The diagnosis is free and asks for no card.