mirror of
https://github.com/Dicklesworthstone/coding_agent_session_search.git
synced 2026-09-22 11:59:25 +08:00
Jeff Emanuel
3375d7f468
Advertise cass_view_batch only with explicit startup --db and route it
through the existing strict decoder, request budget, quota and guarded
Session handler. Reuse the exact single-view coordinate schema for each
batch selector. Preserve concurrent cass_refine and unload/reload behavior.
A tools/call retrieves 1-8 complete windows in input order using one archive
open and one read transaction. The entire batch shares 64 KiB of UTF-8 body
content, a deadline and at most 128 requested message occurrences. Any later
missing, mismatched or oversized window fails without exposing a successful
prefix. The archive snapshot is explicitly not a lexical-freshness proof.
Add five real MCP/storage regressions for conditional discovery, exact RPC
correlation and structured/text evidence, invalid selectors and ignored
notifications, identity-failure recovery, and aggregate NUL-heavy response
bounds. Together with the core landing this adds thirteen tests. Update the
existing opt-in catalog assertion without removing or weakening its checks.
Validation: preparation run 35682484173 produced artifact 10676075012 with
SHA256 6f6cdbb9dc8e0cfd715718e9461ec53c0de29482e0fea1c95d7bacf5d60c4e77.
Verified complete input and edited blob hashes, forward/reverse patch and
changed-path checks, then pinned Rust formatting. Downloaded source matches
both exported blobs; the formatter diff changes only reviewed layout. Core
native test run 35682013290 is still in progress. No completed native test,
full compilation, Clippy, RCH or UBS pass is claimed; local tools absent.
Restore the temporary preparation workflow to its original read-only state.
Retain all concurrent changes through 2f16a029 after checking the three base
blobs. Direct main via GitHub, no PR or force update. Related GH452 bounded
follow-up journey; no broad performance, freshness or bead closure claim.