datatug / datatug-core / 38418501493

100%

push

github

feat(detective): add reviewed AI proposals and shared knowledge contracts (#362)

Detective needs typed, reviewable AI proposals and the same private knowledge transitions in local and hosted adapters. Add provider-neutral contracts backed by DataTug's existing aichat tool/agent loop, with no mutation or execution during proposal generation.

Confirmation rechecks current scope and source access, binds the full proposal and operation to an immutable receipt, reconciles exact retries after a successful mutation, and rejects stale or conflicting requests. Reviewed extraction creates a separately bound Case-edit proposal; private cited excerpts are bounded and full source text is not stored. Shared knowledge validation preserves confirmed-finding evidence and target-specific review handoffs.

Validation: whole-module tests report 100% statement coverage; vet, golangci-lint and tidy checks pass. Adverse tests cover stale snapshots, empty collection roundtrips, revoked replay authority, edited-context limits and digests, and native Anthropic/OpenAI tool schema bodies. Astra review is required before landing. Local/backend/UI adapters and real-provider journeys remain consumer work.

Source commits:

- 2828d6cb0721 Add shared Detective model evaluator and lifecycle
- 0b933b27145f Support exact execution validator correlation
- 91e835a2d254 Harden detective evidence and scenario receipts
- de6afeff4fd0 Preserve child validator failure across step policies
- df9f912344f9 lint: resolve Detective staticcheck findings separately
- 6f2968b95c7b test: cover Detective and legacy query revision boundaries
- 1c8c1700baf2 fix: reject unhashable scoped Detective operations
- 25cefa90fdcf Merge remote-tracking branch 'origin/main' into detective-checks-ai
- 02b9f377f001 feat: add reviewed Detective AI and private knowledge contracts
- c5211456749c fix: reconcile Detective AI receipts and reviewed context edits
- 1c0fd4a1a592 fix: bind Detective AI proposals to bounded r... (continued)

1015 of 0 new or added lines in 0 files covered. (NaN%)

14494 of 14494 relevant lines covered (100.0%)

1.0 hits per line