Seven structural checks on a public repository. Scanned 2026-11-02 23:18 · 5,581 files
80.2/ 100 · risk level Medium
A well-run repository. CI matrix, lockfile, pre-commit hooks and a code owner all present, and test imports reach every source package. The exposure is concentrated in one place: 97 business-logic files over 800 lines. That is a context-truncation risk, not a style complaint.
Scorecard
#
Check
Score
Weight
Verdict
C1
Scope Containment
10/10
18%
pass
C2
False-Green Detection
10/10
18%
pass
C3
Drift Containment
8/10
14%
pass
C4
Memory Integrity
7/10
10%
acceptable
C5
Delivery Gate
8/10
16%
pass
C6
Complexity Control
1/10
12%
needs fixing
C7
Regression Surface
10/10
12%
pass
Evidence
C1 Scope Containment10/10
CI config: .github/workflows
CODEOWNERS present: .github/CODEOWNERS
Pre-commit hook: .pre-commit-config.yaml
CI contains gate/guard/policy semantics: yes
C2 False-Green Detection10/10
Source files 2332 / test files 3451 (test ratio 148.0%)
Assertion density: 10827 assertions across 200 test files (54.1 per file)
CI runs the tests: yes
C3 Drift Containment8/10
Agent constraint files: AGENTS.md
Constraint strength: 12 hard prohibitions found (MUST / NEVER / forbidden)
C4 Memory Integrity7/10
Memory/state files: none found
Project shows agent/LLM characteristics: yes
Note: agent project without cross-session memory files. Within a single session this is not a conflict; it becomes a gap the moment state is expected to persist
Three largest files: tests\models\test_openai_responses.py (20662 lines); tests\models\test_anthropic.py (17555 lines); tests\test_agent.py (15395 lines)
Mean above 400 lines/file: granularity too coarse overall
Risk: Oversized files. The agent's context truncates mid-edit and the regression lands unseen.
C7 Regression Surface10/10
Dependency lock: uv.lock
Source packages 10, reachable by test imports 5 (coverage 50%); not covered: .claude, .github, clai, pydantic_ai_slim, scripts
CI matrix across environments: yes
Priority stack
#
Fix first
Score
Why
1
Complexity Control
1/10
Oversized files. The agent's context truncates mid-edit and the regression lands unseen.
2
Memory Integrity
7/10
below threshold
Want this run on your repository?
A 7-day, human-verified audit: scored report, priority stack, and a recorded walkthrough of your own code. Findings are checked by hand and anything that does not hold up is deleted.