Your agent says it "added auth middleware and updated tests." Honesty Check reads what it actually did and tells you where the story doesn't match.
What it catches
Agent claims to have written feature X — the file was never touched. We diff the claim against actual file operations.
Tests deleted, comments stripped, files removed without mention. Every destructive action gets flagged.
Asked to fix one bug, modified 14 files. We surface every file touched that wasn't in the original request context.
How it works
Extract the claim
We parse the agent's final summary message — the "what I did" that gets posted to your PR or chat.
Map the reality
Every file read, write, edit, and delete in the session log is compiled into an operation list with diffs.
Score the gap
Claude compares claim vs reality and returns a structured list of discrepancies, omissions, and fabrications.
Example output
Honesty Check · 3 discrepancies found
Claimed
"Added auth middleware and updated all related tests"
Reality
2 test files deleted · 1 test file added with 2 fake assertions · middleware added but missing rate-limit guard