Leaderboard github.com
Weizhena/Deep-Research-skills
Structured deep research skill for Claude Code/Open Code/Codex with human-in-the-loop control
unreviewed-ai-risk
AI provenance
66%
explicit
- Confidence
- 82%
- Coverage
- 100%
Estimated share of AI-generated code. Explicit attribution is a lower bound.
Code quality
42/100
poor
- Confidence
- 90%
- Coverage
- 100%
Complexity, duplication, lint density, tests, CI, docs and review process.
Security
82/100
adequate
- Confidence
- 85%
- Coverage
- 100%
Secrets, static analysis, vulnerable dependencies, CI and supply-chain posture.
- Repository: https://github.com/Weizhena/Deep-Research-skills
- Commit:
6ce38f60e3f8b22502c29873f96503a4e0c5addb - Generated: 2026-11-14T00:14:52+00:00
- Scorer 0.1.0, scoring prov-4+qual-2+sec-2, report schema v2
Scores
Unreviewed AI code risk: a large share of this code is explicitly AI-attributed, and its tests or review process are weak while quality or security scores are low.
| Pillar | Score | Signal | Confidence | Coverage |
|---|---|---|---|---|
| AI provenance (estimated AI-generated share) | 66 | explicit | 82% | 100% |
| Code quality | 42 | poor | 90% | 100% |
| Security | 82 | adequate | 85% | 100% |
AI provenance
Explicit AI attribution on 67% of first-party code added; an estimated 67% including velocity signals.
- Explicitly attributed share (lower bound): 66.5%
- Unattributed share in AI-consistent velocity bursts: 0.0%
- Estimated AI-generated share: 66.5%
| AI tool | Attributed commits | Lines added |
|---|---|---|
| Claude Code | 33 | 1,100 |
Current source lines last changed by AI-attributed commits: 388 of 833.
Too few lines in one group to compare finding densities.
| Month | Lines added | Attributed to AI |
|---|---|---|
| 2025-12 | 270 | 100% |
| 2026-01 | 429 | 69% |
| 2026-03 | 423 | 0% |
| 2026-06 | 357 | 100% |
| 2026-08 | 175 | 100% |
Evidence
- Claude Code: trailer 'Co-Authored-By: Claude' (6ce38f60e3f8)
- Claude Code: trailer 'Co-Authored-By: Claude' (e24c5cb46642)
- Claude Code: trailer 'Co-Authored-By: Claude' (dc18cf459144)
- Claude Code: trailer 'Co-Authored-By: Claude' (e932c92cc688)
- Claude Code: trailer 'Co-Authored-By: Claude' (510b6ba1982b)
- Claude Code: trailer 'Co-Authored-By: Claude' (8919ab2e561f)
- Claude Code: trailer 'Co-Authored-By: Claude' (5393c79b534d)
- Claude Code: trailer 'Co-Authored-By: Claude' (7426ee838321)
- Claude Code: trailer 'Co-Authored-By: Claude' (437bb28e69aa)
- Claude Code: trailer 'Co-Authored-By: Claude' (a92028e33bd8)
- Claude Code: trailer 'Co-Authored-By: Claude' (bb7b8453e252)
- Claude Code: trailer 'Co-Authored-By: Claude' (46529e4b88f6)
- Claude Code: trailer 'Co-Authored-By: Claude' (779c44767037)
- Claude Code: trailer 'Co-Authored-By: Claude' (98ddfce42dbe)
- Claude Code: trailer 'Co-Authored-By: Claude' (72cd188d528f)
AI provenance describes how code was produced; it is not a quality judgment. Unless attribution is explicit, these are signals consistent with AI generation, not proof. Squash/rebase workflows, imported code, and rewritten history reduce accuracy.
Code quality
Code quality is poor: test-to-source ratio 0.19, no CI test run detected; 54% of function code in complex or long functions; 82% duplicated lines. Weakest area: duplication.
| Area | Score |
|---|---|
| Complexity (lizard) | 20 |
| Duplication | 1 |
| Code smells (Opengrep) | 100 |
| Tests and CI | 39 |
| Documentation | 65 |
| Type safety | 40 |
| Process (Scorecard review/maintenance) | n/a |
- Functions: 20; mean complexity 12.8, 90th percentile 17; 8 complex and 0 long functions
- Duplicated lines: 82.4% of 682 compared
- Tests: 156 test lines for 833 source lines (ratio 0.19); CI test commands: none detected; 27% of recent source commits also change tests
- Docs missing or thin: contributing, changelog
Most complex functions:
- load_fields_yaml in skills/research-codex-en/research/validate_json.py: complexity 17, 30 lines
- load_fields_yaml in skills/research-codex-zh/research/validate_json.py: complexity 17, 30 lines
- load_fields_yaml in skills/research-en/research/validate_json.py: complexity 17, 30 lines
- load_fields_yaml in skills/research-zh/research/validate_json.py: complexity 17, 30 lines
- main in skills/research-codex-en/research/validate_json.py: complexity 16, 46 lines
Security
Security posture is adequate: no high-impact findings. Weakest area: posture.
| Area | Score |
|---|---|
| Secrets (gitleaks, tree + history) | 100 |
| Static analysis (Opengrep) | 100 |
| Vulnerable dependencies (OSV) | n/a |
| CI workflows (zizmor) | n/a |
| Posture (policy, updates, signing, lockfiles) | 0 |
| OpenSSF Scorecard | n/a |
- Posture checks passed: lockfiles; missing: security policy, dependency updates, signed commits
Static signals only: the scorer never runs the project's code. Secret findings are pattern matches that were not verified, and secret values are never stored.
Repository profile
- Source lines: 833; test lines: 156
| Language | Lines |
|---|---|
| Python | 886 |
| Shell | 103 |