In development: not installable yet

Find what your tests would let through.

Truecount is a GitHub App for pull request review and repository audits. It breaks your changed code on purpose to see whether your tests notice, checks your docs against your code, and reviews pull requests against the guidelines you already wrote.

What a finding looks like

The tests caught 62 of 80 deliberate breaks in the changed lines.

Not caught
- truecount/admin.py:54       not rows → rows                 (remove not)
- truecount/retention.py:75   conn.execute('ROLLBACK') → pass (remove call)
- truecount/settings.py:100   return int(raw) → return None   (return None)

From Truecount's own repository, the day the check was built. Each line is a change the test suite would have accepted. The missing tests were written, and the next run caught 93 of 102. Some breaks change nothing observable, so a survivor is a lead to check, not a verdict.

The agents

Agent	Runs when	Runs on	Can change
Pull request review	A collaborator opens a pull request, or a maintainer asks	Truecount (reads the diff)	Nothing; it comments
Mutation check	On pull requests (Python)	Your Actions runner	Nothing; it reports
Docs audit	Weekly, skipped if the code has not changed	Your Actions runner	Markdown and `docs/`, through a pull request
Code health	Weekly, skipped if the code has not changed	Truecount reports; your runner fixes	One fix per run, through a pull request
Bug reproduction	You label an issue	Your Actions runner	Nothing; it comments with what it ran

Configure it

# .github/truecount.yml, read from your default branch
review:
  enabled: true
  guidelines: .github/review-guidelines.md
ignore:
  - "docs/generated/**"

Ask for a review with a /truecount review comment or the truecount-review label. The config is read from the branch a pull request targets, so a pull request cannot change how it is reviewed.

Guarantees

Your code only executes in your own Actions runner.
No agent holds a token that can write to your repository. Agents send changes to Truecount, which checks every path against that agent's rules before opening a pull request.
No agent can edit workflows, its own config, or your review guidelines.
Nothing merges itself.
Every file a review skips is listed in the review.
Your code is never used to train models.

Plans

Plan	For	Includes
Free	Public repositories	Pull request review, docs audit and code health, with a monthly run allowance
Individual	Personal accounts	Private repositories and every agent, with a generous run allowance
Team	Organizations	Private repositories and every agent, with an allowance sized for a team
Enterprise	Large organizations	Runs on your own model endpoint, under an annual license

A run is one review or one audit. Runs past your allowance wait for spare capacity instead of failing, and top-ups are available. Prices and allowances will be published before launch.