All issues

Search results

Enhance resume accounting counts raw checkpoint-dir ids (dataset membership never checked) and never retires the errors it seeds — Done: 1825/1819, phase=done errors: 528 with zero errored checkpoints on disk

Status: Open.

#748 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

resolvePackageCall matches cross-package calls on the import's directory basename over a map-ordered candidate list — wrong-package edges, different across parses of the identical tree (21 edge flips; 107 units inlined a different callee body)

Status: Open.

#747 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

The Go call graph records no edge for method values held in struct-literal fields or named-func-type containers — the OpenBao handler idiom: 254 of 318 Callback handlers pruned with no caller in one scan (the AppRole login handler among them)

Status: Open.

#746 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

A delegation-dismissed SAFE row names its sink-owner only in prose — no consumer surface carries the edge, and when the sink-owner was pruned the finding evaporates (one executed chain; Stage 2 wrote 'a separate sink, not the target function' and nothing chased it)

Status: Open.

#745 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

The Stage-1 verdict vocabulary has no definitions in the prompt and INCONCLUSIVE's disposition contradicts itself across surfaces — manual-review-recommended (docs, HTML legend, remediation) while disclosure-dropped and never Stage-2-verified; a semantics ruling

Status: Open.

#744 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

A SAFE rationale can rest on a factually false claim about the code — the cassandra username template 'strips CQL-special chars' but no transform strips the quote character — and Stage 2 never re-examines SAFE rows

Status: Open.

#743 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

An all-incomplete enhancement phase still rolls up as status success — incomplete_count never derives a step status

core:scanner

Status: Open.

#742 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

The orphan-rate advisory divides by pruned units — in library mode it makes coverage look worse the better it works

enhancer

Status: Open.

#741 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

Enhance exception-path spend never reaches AgentResult.cost_usd — per-unit costs under-count ~8-10% and the tracker and the report disagree

bugSomething isn't workingSomething isn't working

enhancer

Status: Open.

#740 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

--limit N selects units by file-walk order, not analysis priority — a limited run silently measures the least interesting code

bugSomething isn't workingSomething isn't working

core:scanner

Status: Open.

#739 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

--library-mode is not auto-applied (or suggested) when app-context classifies the target as a library — the classification arrives after the filter already pruned the public API

bugSomething isn't workingSomething isn't working

core:scanner

Status: Open.

#738 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026

Stage-1 confidence is a free 0-1 field with no rubric and no decision consumer, yet it ships to the report and CSV as stage1_confidence — across 213 Stage-2-verified findings it barely separates agreed from refuted (AUC 0.57) — define it, consume it, or drop it

Status: Open.

#736 In knostic/OpenAnt;

· gadievron opened on Sep 25, 2026