A record of an investigation that used hegel-rust property-based tests — written by subagents driving the hegel-skill — to find real, upstream-reportable bugs in a wide range of open-source Rust crates, and to develop the hegel-skill from what the runs revealed.
runs.md, pinned commits per crate), full per-run evaluations (batch2-eval.md), per-candidate disposition (coverage.md), the extended-candidate selection heuristic (extended-candidates.md), and ready-to-file issue drafts (draft-issues.md).git apply-able .patch per project. PATCHES.md records, for every patch, the upstream repository and the exact base commit (SHA, date, subject) the patch was written and verified against. Crates whose maintainers opted out of AI contributions are omitted.Every suspected bug was recorded only after: an independent deterministic reproduction against pristine upstream; confirmation it violates a documented or self-evident contract (oracle-independent failures — panics, aborts, invariant violations, self-roundtrip failures — weighted above differential disagreements, especially in hardened crates); a check against the upstream issue tracker for duplicates; and exclusion of crates whose maintainers have opted out of AI contributions.
Nothing has been filed to any third-party repository — the findings are staged for review. Crates whose maintainers opted out of AI contributions were excluded and are noted as such. The actual per-crate property tests and HEGEL_REPORT.md files (under targets/, not committed here) remain local.
92 commits
Hacker News (1)
Rust
97.4%
Python
2.6%
A record of an investigation that used hegel-rust property-based tests — written by subagents driving the hegel-skill — to find real, upstream-reportable bugs in a wide range of open-source Rust crates, and to develop the hegel-skill from what the runs revealed.
runs.md, pinned commits per crate), full per-run evaluations (batch2-eval.md), per-candidate disposition (coverage.md), the extended-candidate selection heuristic (extended-candidates.md), and ready-to-file issue drafts (draft-issues.md).git apply-able .patch per project. PATCHES.md records, for every patch, the upstream repository and the exact base commit (SHA, date, subject) the patch was written and verified against. Crates whose maintainers opted out of AI contributions are omitted.Every suspected bug was recorded only after: an independent deterministic reproduction against pristine upstream; confirmation it violates a documented or self-evident contract (oracle-independent failures — panics, aborts, invariant violations, self-roundtrip failures — weighted above differential disagreements, especially in hardened crates); a check against the upstream issue tracker for duplicates; and exclusion of crates whose maintainers have opted out of AI contributions.
Nothing has been filed to any third-party repository — the findings are staged for review. Crates whose maintainers opted out of AI contributions were excluded and are noted as such. The actual per-crate property tests and HEGEL_REPORT.md files (under targets/, not committed here) remain local.
Hacker News (1)
92 commits
Rust
97.4%
Python
2.6%