feat: TMCRA 1.0.0-rc.1 local workspace (scanner adjudication pending)

[reshuibuduo]

Summary

Unified TMCRA 1.0.0-rc.1 candidate: memory workspace, local Writer/organizer settings, knowledge/graph views, session modes, task continuity, bounded evidence budgets, and interactive chat confirmation for exact-source corrections.

Account-free Windows x64 installation bundles the actual backend and verified inventory. Install-Local.cmd / tmcra_open_local_install prepare private Python, pinned models and a local identity. Setup has no API credential and exposes no memory actions; after installation the host must restart. First-time downloads need internet, while memory runtime does not need TMCRA servers/accounts.

Verification

The Codex and Claude Code contract suites pass, including 15 lifecycle scenarios, 7 authorization flows, confirmation/no-write-before-consent, local/cloud isolation and account-free setup. The 235-entry release archive and binary/runtime integrity checks pass. Latest code tests: https://github.com/reshuibuduo/tmcra-plugin-codex/actions/runs/33988900575

Publication blocked by scanner

The official HOL/Cisco job still fails: https://github.com/reshuibuduo/tmcra-plugin-codex/actions/runs/33988900546
The native detector misclassifies PyTorch Module.eval() in eight bundled Python files. The current Action additionally flags synthetic test-fixture credentials because repository test exclusions are not trusted by default. Local CLI and raw CI reports therefore differ; neither is a passing severity gate. No new exclusions, baseline suppressions, or threshold reductions were introduced.

Reviewed source paths and limitations: https://github.com/reshuibuduo/tmcra-plugin-codex/blob/codex/memory-controls-continuity/docs/security-scanner-review.md

This PR and three dependent Codex/Claude catalog refreshes remain draft pending an upstream detector fix or maintainer adjudication. No standalone Codex 1.0.0-rc.1 tag/release has been published. Main TMCRA runtime, DSH and generic MCP are separate 1.0.0-rc.1 releases.

Local-model acceptance is partial

Synthetic CPU ingest/raw recall passed; complex compilation timed out at 600 seconds. Organizer/full-service restart, latest memory-limit settings and balanced/high-tier hardware acceptance remain pending. Current laptop lacks approximately 6.3 GiB of free RAM for the next full run. A cloud-hosted Agent main model may still receive recalled evidence. Production services, real user memory and currently installed plugins were not changed.

[reshuibuduo]

Upstream detector fix submitted: hashgraph-online/hol-guard#2805

The patch removes the eight Python inference-method false positives on the unchanged bundled runtime. Local validation: 186 related tests passed on Python 3.12; 93 code-quality tests passed on Python 3.10. Genuine dynamic-evaluation and JavaScript/TypeScript cases retain high-severity findings.

Upstream GitHub Actions currently require maintainer approval. This release stays a draft: the official pinned Action is unchanged, the score-80/high-severity gate and Cisco scanning remain in place, and synthetic test-credential findings still need separate resolution. This local regression result is not a passing official marketplace scan.