codex-config-portable

Run and recover long-lived Codex CLI conversations in Linux tmux sessions with workspace-scoped metadata and SESSION_ID tracking. Use when Codex needs to keep watching one interactive session, continue the same conversation after detach/reattach, recover after terminal loss with exact `SESSION_ID` resume, or monitor tmux-backed Codex output without relying on VS Code integrated terminals.

Build, edit, render, import, and export presentation decks with the preloaded @oai/artifact-tool JavaScript surface through the artifacts tool.

Build, edit, recalculate, import, and export spreadsheet workbooks with the preloaded @oai/artifact-tool JavaScript surface through the artifacts tool.

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

Use when you have a written implementation plan to execute in a separate session with review checkpoints

Check Codex "cloud review" signals on a GitHub PR (thumbs-up reactions, comments/reviews, checks) and troubleshoot cases where the review appears to have "no feedback". Use when a PR is expected to be reviewed by Codex but gh/CLI output shows nothing.

Implements Manus-style file-based planning to organize and track progress on complex tasks. Creates task_plan.md, findings.md, and progress.md. Use when asked to plan out, break down, or organize a multi-step project, research task, or any work requiring >5 tool calls. Supports automatic session recovery after /clear.

Use when executing implementation plans with independent tasks in the current session

Use when starting any conversation to enforce strict skill-first workflow discipline in Codex, including invoking relevant skills before actions and keeping execution checklists synced via update_plan.

Export paper-ready artifacts (figures/tables/metrics/config/README and small tex snippets) from an experiment output directory into ~/project/exp/paper_exports/{project}/{run_id}/ for local incremental pulling via paper-sync. Use when you need a stable, repeatable export layout with a manifest.json (timestamp, run_id, git commit if available, and a best-effort metric summary). Safe by default (no destructive deletes; skips large training artifacts, checkpoints, and common binary blobs).

Manage iterative artifacts by keeping a canonical "latest" file in the repo while archiving snapshots to WORKSPACE/.codex/tmp/artifacts/ARTIFACT_ID/history with a small manifest. Use when repeated revisions would otherwise create many versioned files (png/pdf/svg/yaml/logs/scripts) and you need clear latest vs history without clutter. Never delete without explicit user approval.

Create, edit, and review diagrams.net (draw.io) files. Use for .drawio XML edits, layout/coordinate adjustments, exporting .drawio to PNG/SVG/PDF, and AWS icon lookups (mxgraph.aws4.*).

Check idle GPU availability across both nodes (gpu01 and gpu02) and recommend candidate devices before launching jobs. Use when you need to choose a node/GPU for training or inference, or to report GPU availability with a concrete recommendation.

Launch and monitor long-running commands with resumable status + logs under WORKSPACE/.codex/longrun, and optionally chain follow-up steps after completion. Use when a job will likely exceed a reasonable single-session window, or when you want the agent to keep progressing without manual reminders.

Inventory untracked/temporary artifacts in the active workspace and produce a safe cleanup plan (keep/temporary/review) with suggested move/delete commands. Use when a task created many intermediate files (tmp scripts, duplicate YAMLs, logs, outputs) and you want to avoid workspace clutter. Do not delete anything without explicit user approval.