ift-ts:qa:logos:2026q4-source-code-level-qa
Description
Build an AI code analysis system prototype that helps discover new bugs quickly and turns findings into actionable issues. Use a second agent to independently review candidate findings and reduce false positives, then provide precise classifications and fix proposals. Generate proofs or exploit reproductions for high-severity issues and regression tests where possible.
Task List
System prototype
- fully qualified name:
ift-ts:qa:logos:2026q4-source-code-level-qa:system-prototype - owner: Roman
- status: not started
- start-date: 2026/10/01
- end-date: 2026/12/31
Description
Prototype an AI code analysis workflow covering:
- Fast discovery of new bugs through agent-assisted code analysis.
- Independent cross-review by a second agent to challenge findings and reduce false positives.
- Precise issue classification, including impact, severity, affected scope, supporting code evidence, and a proposed fix. Use LEZ issue #866 as an example of the expected reporting detail.
- Proof or exploit generation to reproduce and demonstrate high-severity findings.
- Regression test creation where possible, so confirmed bugs can be detected automatically in future changes.
Deliverables
- A working prototype and documentation for the discovery, cross-review, and reporting workflow.
- Classified findings with supporting evidence, second-agent review results, and fix proposals.
- Reproducible proofs or exploits for high-severity findings, with any reproduction limitations recorded.
- Regression test PRs where feasible, with remaining test coverage gaps documented.