Skip to content

silent9669/codex-atlas

v0.5.2

Repository graph workspace, document organization, audit evidence and local model decisions for Codex.

Changelog

0.5.2

  • Disable inherited tool access during background Codex inference while retaining the configured provider and authentication.

0.5.1

  • Unify repository identity, navigation, Commands and Assistant in one desktop header.
  • Use a compact second navigation row on narrow screens, with accessible icon controls on phones.
  • Remove the duplicate overview toolbar and preserve document-scoped chat and drawer focus.

0.5.0

  • Add a compact Overview with document categories, review queue, focused connections and latest execution.
  • Add graph relationship filters, neighbor highlighting and focused views that preserve the selected file.
  • Provide direct atlas-organize, atlas-audit, atlas-docs, atlas-memory and atlas-graph skills alongside the same app workflows.
  • Support workspace URL fragments and preserve keyboard focus when closing workflow drawers.

0.4.2

  • Improve reading with 17px text, a 700px column, larger tables and bundled Vietnamese glyphs.
  • Simplify document actions and metadata; add a heading navigator that adapts to narrow screens.
  • Make category navigation clearer with larger headings, visible counts, selected-document expansion and expand/collapse controls.
  • Restore reader interactivity after mobile library selection.

0.4.1

  • Compact workspace navigation into a top bar and give the document library more vertical space.
  • Combine document labels and actions, tighten reader spacing, and use full-width Graph, Runs, Memory and Audit views.
  • Preserve mobile Library access, keyboard focus and reduced-motion drawer behavior.

0.4.0

  • Replace the document table with a category library and safe Markdown reader; keep chat beside the selected document.
  • Add Runs and Memory views, source-bound relevant recall, directional graph evidence and live execution stages.
  • Auto-load ignored repository-root .env.local; preserve shell precedence and saved CLI authentication.
  • Validate against an isolated MegaDay_Twin clone; install the supported Codex compatibility prompt.

0.2.0 — 2026-10-31

  • Added one live graph page with responsive document, audit, decision and chat drawers, keyboard navigation and reduced motion.
  • Added the canonical $codex-atlas skill, optional CLI prompt, automatic workspace startup and small-model routing to separate Codex execution.
  • Added automatic document labels, topic/status filters, reviewed overrides and stale-source detection.
  • Added idempotent project documentation and memory updates after successful implementation runs.
  • Added local document classification, Vietnamese acceptance signals, extraction-audit findings, and bounded Codex handoffs.
  • Added explicit model routing, optional OpenAI Responses API discovery/advice, strict cited-path validation, local env-file loading, capped output, usage records, and refusal/secret handling.
  • Added safe Codex child launching with explicit model, reasoning effort, schema, sandbox, and stale-result checks.
  • Validated against the public silent9669/MegaDay_Twin snapshot: 4,234 nodes, 5,834 relationships, 52 documents, and 10 extraction findings.

0.1.0 — 2026-10-31

  • Local Markdown, Python AST, TypeScript compiler AST and Git graph extraction.
  • Cited context, bounded structural impact, explanation, changed-file review and JSON reports.
  • Reviewed check execution with sanitized artifacts, scoped evidence links and stale snapshot detection.
  • Searchable keyboard-accessible static HTML report.
  • Portable Codex plugin and Git-backed marketplace installation.

This release does not implement MegaDay's separate city-chat grounding integration or establish live model quality.

0.3.0

  • Document-first workspace with Documents/Graph navigation, optional filters, one assistant/detail panel, and concise action selection.
  • Saved Codex authentication for structured chat, category and routing calls; explicit API keys retain direct Responses behavior.
  • AI-created document categories, selected-document questions, docs/audit/memory commands, and reviewed-label precedence.
  • Explicit completion outcomes prevent failed child checks from creating success notes or memory.
  • Clean local marketplace packaging excludes runtime files and Git history.
  • Local marketplace installation and live model/CLI validation; nested macOS sandbox restrictions are reported without silently falling back.