Repository navigation
feat(lcm): recover evidence and search external history - #52
Merged
Merged
Conversation
|
ScriptedAlchemy
changed the title
Co-Authored-By: GPT-6 Astra <noreply@openai.com>
…very Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
main removed effect/Crypto, so the new history services and their tests move to Digest and NodeCrypto, and the toolkit test uses the in-process MCP layer. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…der switch tests Provider handoff records now keep the compact coverage receipt in summaryText and the intact evidence in history.messages; the delivered provider text is rendered from history. Three ProviderSwitch cases still read the transcript from summaryText or the handoff activity item. The compact-legacy and legacy-replacement fixtures stripped history from a new record, which produced a receipt-only record without history that no server version writes. They now model a pre-history record with the rendered transcript in summaryText, so the inline-delivery assertion stays unchanged. The queued-Claude and imported-v1 cases now assert the item summary equals the receipt, that the receipt omits the transcript, and that history.messages carries the exact messages, alongside the unchanged provider-text assertions. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The tool reads raw Codex homes across projects, including sessions outside T3, which is broader than a delegated child needs. Also record that summaries cannot be rolled back to an older server. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # packages/contracts/src/rpc.ts
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
T3 could offer incomplete or stale conversation-summary coverage, and a bounded handoff could lose the exact evidence needed to verify its summary. Handoffs also persisted a duplicate rendered transcript. Conversations created directly in Codex or exported from ChatGPT were unavailable to T3's history tools.
Changes
t3_external_historyMCP tool. Native Codex reads use enabled providers' configured shared Codex home, validate that home at handshake, deduplicate providers sharing it, and retry another reader only for unavailable/unsupported responses.Validation
git diff --check.--dry-run. No paid model evaluation or live authenticated Codex archive was used.Context budget and limits
The new MCP tool intentionally increases the measured tool-list JSON from 64,883 to 66,169 bytes (+1,286); the provider instruction budgets are unchanged. The committed budget matches the actual catalog.
External search is literal and paged. ChatGPT support reads configured exports, not a live cloud sync; reaching later exported conversations currently rescans the file prefix. Native evidence contains normalized visible text, diagnostic output and diffs rather than a raw rollout JSON dump.
T3 retains leaf summaries and bounded offered handoff groups. This does not claim hierarchical-compaction or archive-scale quality/performance parity with TraceDecay.
The repository's
test-t3-appandtest-t3-mobileskills require managed Browser/Device panel tools unavailable in this session. Automated component/browser checks passed, but live integrated client/device validation and before/after captures remain outstanding, so this PR remains draft. Live model evals also remain outstanding because no evaluation credentials are configured.Implemented and reviewed with GPT-6 Astra Pro in ChatGPT.