FALSE FLOORS / FREE TRUST-CHECK SKILL

Sick of cleaning up after your agent?

You’ve already added rules for the mistakes you’ve seen. You shouldn’t have to discover every failure the hard way.

Find out now what else can go wrong, which controls you already have, and what you could add.

Check my coverage
Run the free False Floors check against your repository and agent setup.
REAL STAGE 1 RESULT

1 supported component found

VERIFIED PARTIAL
RL-2C — destructive-command guard installed; known-bad self-test passed.
52 PATTERNS NOT FOUND
Stage 1 detects configuration, not runtime effectiveness.
STAGE 2 QUEUE
1 wired control identified for behavioral testing.
Real local Stage 1 result — 11 September 2026. No score or certification.
THE FALSE FLOORS CATALOGUE

A free, regularly updated catalogue of every coding-agent failure we can find, matched to the controls available to prevent or catch it.

New failures and controls will keep emerging. The catalogue records what we can support with evidence today. When a failure has no reliable technical control, it says so.

01 / UNDERSTAND YOUR EXPOSURE

How much does your setup cover?

You probably have more controls than you think. Tests and repository safeguards can catch an agent before a mistake reaches production.

Checking your coverage is like reviewing your insurance. You want to know which failures have something in place to catch them, and where you are still exposed.

But most controls arrive after something has already gone wrong:

• Your agent skips a test.
• It edits the wrong file.
• It repeats a mistake you thought you had fixed.

You add another rule and hope that closes it. False Floors compares your repository and agent configuration with the catalogue. It shows your coverage and the evidence it found.

02 / INSPECT THE FINDINGS

Can you check the findings yourself?

The check looks at your configuration and quotes the evidence it finds. You can trace each finding back to the control in your setup.

A remembered “yes, we have branch protection” isn’t enough. The check looks for evidence that the control is present. When it can’t establish an answer, the result stays unknown.

Coverage means there is evidence of a relevant control in your setup. It doesn’t prove that the control will catch the failure under pressure. That takes a separate test.

What do you get back?

You get a coverage report and a machine-readable record for later comparisons.

01
See the controls already present and the evidence behind each finding.
02
Find missing controls and the failures they could address.
03
Investigate unknowns before treating them as covered.
04
Keep the decisions that still need human judgement with a person.
03 / RUN IT WHEN YOU NEED IT

How much work does it add?

Run the check when you need it. It doesn’t remain in your workflow or run in the background.

Local scripts handle the routine inspection. Your coding assistant explains the results and writes the report. The catalogue stays on disk, so the assistant doesn’t need to read every entry into the conversation.

04 / KEEP YOUR COVERAGE CURRENT

How do you stay current?

Every report records the catalogue version used for the check. Re-run it after an update to see the failures and controls added since your last review.

That gives you a repeatable way to review coverage as your agent setup changes and the field learns from new failures.

FOR ENGINEERING MANAGERS AND CIOs

What can managers and CIOs ask their teams to show them?

If an agent can report passing tests without running them, you need to know which repositories have a control for that failure.

Ask each team to run the same check. Review the evidence together and agree which gaps to address first. Give every unresolved judgement an owner.

The reports give you a common list of failures and controls across repositories. Keep the dated results so each review can show what changed in the team’s setup and in the catalogue.

The check assesses the presence of controls for catalogued failures. It doesn’t provide certification.

HELP EXPAND THE CATALOGUE

Found a failure we’ve missed?

Tell us what happened. We’ll investigate what could prevent or catch it and publish what we find. Reporting it could save the next developer from discovering the same failure the hard way.

Submit a failure
START WITH ONE REPOSITORY

Check your coverage before your next agent task.

Start with one repository. Choose the next gap to address.

Check my coverage
Code: MIT. Catalogue content: CC BY 4.0.