Recovery/audit 20260529

06_PHASE_PROMPTS/PHASE_00_PREFLIGHT_AND_BASELINE.md

@@ -0,0 +1,13 @@
+# Phase 00 — Preflight and Baseline
+
+Run the preflight commands from the master prompt. Produce `codex/receipts/startup_preflight.json` and `codex/reports/phase_00_report.md`.
+
+Acceptance:
+- current repo root identified;
+- git state recorded;
+- tool availability recorded;
+- current run truth surfaces inventoried;
+- known build/test commands listed;
+- no edits made before report.
+
+Stop if repo root or package scope is ambiguous.

06_PHASE_PROMPTS/PHASE_01_CHAT_STREAM_TERMINATION.md

@@ -0,0 +1,15 @@
+# Phase 01 — Chat Stream Termination
+
+Fix the core chat infinite-response bug.
+
+Tasks:
+1. Inspect `src-tauri/src/commands/chat/mod.rs` and provider stream code.
+2. Terminate on provider `done=true`, not HTTP EOF.
+3. Emit exactly one terminal event.
+4. Preserve final-frame metadata.
+5. Add done-without-EOF test.
+
+Acceptance:
+- mock stream with `done=true` and no EOF completes within one second;
+- receipt records `done_frame_seen=true`, `terminal_cause=provider_done_frame`, `eof_seen=false`;
+- frontend leaves streaming state.

06_PHASE_PROMPTS/PHASE_02_PARTIAL_PERSISTENCE_AND_CANCELLATION.md

@@ -0,0 +1,13 @@
+# Phase 02 — Partial Persistence and Cancellation
+
+Tasks:
+1. Add generation attempt/lifecycle persistence.
+2. Persist partial content on timeout/error/cancel.
+3. Make `stop_chat` backend-authoritative.
+4. Remove frontend-local durable assistant creation on stop.
+5. Add continuation lineage.
+
+Acceptance:
+- token then idle timeout creates persisted partial + receipt;
+- stop creates one backend-sourced cancelled assistant state;
+- continuation links to partial attempt.

06_PHASE_PROMPTS/PHASE_03_RUNTIME_GATES_AND_BACKGROUND_PREEMPTION.md

@@ -0,0 +1,12 @@
+# Phase 03 — Runtime Gates and Background Preemption
+
+Tasks:
+1. Audit GPU/LLM gate owners and background summary jobs.
+2. Foreground chat must preempt/cancel/degrade background LLM work.
+3. Add gate-owner/status receipts.
+4. Add UI phase state for waiting/provider-starting/streaming/stale.
+
+Acceptance:
+- blocked background summary cannot make chat appear infinite;
+- chat starts or reports clear bounded status within SLO;
+- all gate waits are receipted.

06_PHASE_PROMPTS/PHASE_04_VALIDATION_GATE_HARDENING.md

@@ -0,0 +1,13 @@
+# Phase 04 — Validation Gate Hardening
+
+Tasks:
+1. Fix package scope gate.
+2. Include/rename missing secret-store permissions gate.
+3. Add subgate timeouts to release candidate gate.
+4. Fail release gate on `release_blocker=true`.
+5. Strengthen static chat audits.
+
+Acceptance:
+- every gate emits structured JSON;
+- aggregate release gate never hangs;
+- current known blockers fail gates until fixed.

06_PHASE_PROMPTS/PHASE_05_PACKAGE_RUN_TRUTH_AND_TRANSFERABILITY.md

@@ -0,0 +1,13 @@
+# Phase 05 — Package, Run Truth, Transferability
+
+Tasks:
+1. Add single RunTruthV1 source.
+2. Add PackManifestV1 inside package root.
+3. Add GitStateReceiptV1 and ToolAvailabilityReceiptV1.
+4. Separate Codex context package from release/source package.
+5. Preserve required receipts and sanitized evidence summaries.
+
+Acceptance:
+- fresh unzip package gate passes;
+- no run truth drift across CURRENT_RUN, sidecars, final handoff;
+- package contains no unrelated root docs in release mode.

06_PHASE_PROMPTS/PHASE_06_RELEASE_PROOF_DESKTOP_INSTALLER_TQ.md

@@ -0,0 +1,13 @@
+# Phase 06 — Release Proof
+
+Tasks:
+1. Run Rust fmt/check/test/clippy.
+2. Run npm install/build/test.
+3. Run live or mock Ollama chat smoke, including done-frame proof.
+4. Run live desktop smoke or mark release blocked.
+5. Produce semantic-memory and TurboQuant runtime receipts.
+6. Run installer smoke or demote installer claims.
+
+Acceptance:
+- no S0/S1 proof gap remains;
+- every unavailable live check is explicitly a release blocker or non-goal.

06_PHASE_PROMPTS/PHASE_07_FINAL_AUDIT_AND_CLAIM_BOUNDARY.md

@@ -0,0 +1,12 @@
+# Phase 07 — Final Audit and Claim Boundary
+
+Tasks:
+1. Generate release proof packet.
+2. Generate public claim diff.
+3. Generate remaining delta.
+4. Run all final validation commands.
+5. Populate hostile auditor handoff.
+
+Acceptance:
+- final handoff includes changed files, commands, pass/fail/skip reasons, blockers, risks, rollback;
+- no release/public claim contradicts receipts.

AGENTS.md

@@ -1,103 +1,46 @@
-# AGENTS.md — Gloss Release-Proof Repair
+# AGENTS.md — Gloss Closing Pass Rules
 
-## Project
+Active run: `GLOSS_TOTAL_COMPLETION_AND_HARDENING_SUPERPASS_20260526`
 
-Gloss is a local-first Tauri + React + Rust notebook/RAG desktop app. The current task is a release-proof repair pass, not a feature expansion.
+## Project purpose
 
-## Current run
+Gloss is a local-first notebook/RAG/chat application. Chat must work without retrieval when retrieval/source state is degraded. Retrieval, semantic-memory, TurboQuant, and release claims must be proof-bearing.
 
-`GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525`
+## Source-of-truth ownership
 
-Write all run receipts/reports under:
+- Frontend chat lifecycle: `src/stores/chatStore.ts`, `src/components/chat/ChatPanel.tsx`, `src/App.tsx` event forwarding.
+- Backend chat lifecycle: `src-tauri/src/commands/chat/mod.rs`.
+- Provider validation and model registry: `src-tauri/src/providers/*`, settings commands, settings UI.
+- Source/retrieval selection: `src/stores/sourceStore.ts`, retrieval backend modules.
+- Package/release proof: `scripts/`, `validation/`, `docs/codex-runs/*`, package sidecars.
 
-```text
-docs/codex-runs/GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525/
-```
-
-Historical run material is evidence only. It must not act as active instruction.
-
-## Source hierarchy
-
-1. Current repo source files, manifests, tests, receipts, and screenshots.
-2. Current run bundle under `docs/codex-runs/GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525/`.
-3. Existing product docs only when consistent with current source.
-4. Historical Codex run docs only as evidence, never as active instruction.
-5. README claims are not proof.
-
-## Hard rules
-
-- Do not add broad features.
-- Do not weaken or bypass release gates.
-- Do not claim release readiness without live receipts.
-- Do not leave old pass artifacts active at repo root.
-- Do not keep P31/P32/P33/P34/P35/P36 instructions as active guidance.
-- Do not silently fallback from semantic-memory/TurboQuant to local retrieval.
-- Do not keep native dense embeddings and semantic-memory embeddings as uncoordinated source-of-truth paths.
-- Do not report pending queue items as running work.
-- Do not report folder import complete unless terminal receipts count ready, failed, skipped, and cancelled states.
-- Do not make vector indexes, caches, UI state, or generated docs source of truth.
+## Forbidden behavior
 
-## Canonical owners
-
-| Concept | Owner |
-|---|---|
-| Product app workflow | `Gloss/src-tauri/src/*`, `Gloss/src/*` |
-| Native dense indexing | `src-tauri/src/ingestion/embed.rs`, `src-tauri/src/state.rs`, notebook DB chunk embedding fields |
-| semantic-memory projection/search | `Libraries/semantic-memory`, with app adapter in `src-tauri/src/memory/semantic_memory_adapter.rs` |
-| Embedding provider boundary | Must be a single app-level provider abstraction used by native dense and semantic-memory projection |
-| TurboQuant candidate artifacts | `Libraries/semantic-memory` + `Libraries/turbo-quant`; Gloss only routes/configures/receipts |
-| Release proof | `docs/codex-runs/GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525/FINAL_RECEIPT.json` and validation scripts |
-| Active Codex instruction | This `AGENTS.md` + current run prompts only |
-
-## Required final artifacts
-
-```text
-docs/codex-runs/GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525/
-  STARTUP_PREFLIGHT.md
-  STALE_PASS_CLEANUP_MANIFEST.json
-  CHANGED_FILES.txt
-  COMMANDS_RUN.log
-  VALIDATION_RESULTS.md
-  EMBEDDING_PROVIDER_RECEIPT.json
-  EMBEDDING_DIAGNOSTICS_RECEIPT.json
-  DENSE_INDEXING_RECEIPT.json
-  SEMANTIC_MEMORY_PROJECTION_RECEIPT.json
-  TURBOQUANT_RUNTIME_RECEIPT.json
-  LIVE_DESKTOP_SMOKE_RECEIPT.json
-  PACKAGE_WARNING_REVIEW.md
-  PUBLIC_CLAIM_DIFF.md
-  FINAL_AUDITOR_HANDOFF.md
-  FINAL_RECEIPT.json
-  REMAINING_DELTA.md
-  ROLLBACK_PLAN.md
-```
+- Do not hide provider errors behind a spinner.
+- Do not block no-retrieval chat because source list loading/partial/error.
+- Do not emit `chat:done` before assistant persistence succeeds unless explicit partial/cancel artifact is emitted.
+- Do not silently allow LAN or cloud endpoints; provider authority must be explicit.
+- Do not claim semantic-memory/TurboQuant/dense indexing proof from dependencies alone.
+- Do not add compatibility shims or shadow truth stores.
+- Do not update release docs by hand if gate JSON says failed.
 
 ## Required validation
 
-Run and record results for:
+Run targeted tests first, then full gates:
 
 ```bash
 npm run build
-node scripts/run_frontend_contract_tests.mjs
-python3 scripts/check_gloss_sm_tq_fix.py .
-python3 scripts/gloss_current_run_truth_gate.py --repo .
-python3 scripts/gloss_receipt_integrity_gate.py --repo .
-python3 scripts/gloss_feature_matrix_gate.py --repo .
-python3 scripts/gloss_issue_ledger_gate.py --ledger ISSUE_LEDGER.csv
-python3 scripts/gloss_static_runtime_truth_gate.py --repo .
-python3 scripts/audit_tauri_security.py .
-python3 scripts/audit_ui_disclosure_a11y.py .
-python3 scripts/gloss_evidence_ui_gate.py --repo .
-python3 scripts/gloss_retrieval_gate.py --repo .
-python3 scripts/gloss_validator_path_gate.py --repo .
-python3 validation/gloss_stale_pass_surface_gate.py --repo .
-python3 validation/gloss_embedding_provider_gate.py --repo .
-python3 validation/gloss_live_receipt_gate.py --repo . --run-id GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525
-python3 validation/gloss_next_release_gate.py --repo . --run-id GLOSS_RELEASE_PROOF_EMBEDDING_UNIFICATION_CLEANUP_20260525
-cargo fmt --manifest-path src-tauri/Cargo.toml -- --check
-cargo test --manifest-path src-tauri/Cargo.toml --features semantic-memory-turbo-quant
-cargo clippy --manifest-path src-tauri/Cargo.toml --features semantic-memory-turbo-quant --all-targets -- -D warnings
-npm run tauri:build:release
+npm test
+cargo fmt --all -- --check
+cargo test --manifest-path src-tauri/Cargo.toml --features semantic-memory-turbo-quant commands::chat::tests
+cargo test --manifest-path src-tauri/Cargo.toml --features semantic-memory-turbo-quant providers::tests
+python3 validation/validate_source_send_gate.py .
+python3 validation/validate_frontend_event_routing.py .
+python3 validation/validate_chat_terminal_contract.py .
+python3 validation/validate_provider_lan_policy.py .
+python3 validation/validate_release_receipt_consistency.py .
 ```
 
-Skipped checks require exact reason and must be listed in `FINAL_RECEIPT.json`.
+## Completion rule
+
+Receipts or it did not happen. Final answer must include hostile-auditor handoff.