Feature request: show Skills, Plugins, and Tools used in each task sidebar

Summary

Please add a thread-scoped Used in this task section to the Codex desktop app's right sidebar, alongside the existing Outputs and Sources sections.

The section should show which Skills, Plugins, and Tools were actually used in that specific task—not merely which capabilities were available.

Motivation

When working across many tasks and projects, it is difficult to remember:

• which Skill supplied the workflow
• which Plugin or connected app supplied a capability
• which Tools were called
• whether a Skill was explicitly selected or activated automatically
• which capability produced a particular result

This information is useful for understanding and reproducing successful workflows, troubleshooting unexpected behavior, auditing external actions, and deciding which Skills or Plugins are genuinely valuable.

Today, recovering this information may require reading the entire conversation and expanding individual activity entries.

Suggested UX

Used in this task

Skills (3)
  skill-router                 Automatically selected
  openai-docs                 Automatically selected
  github                      Explicitly selected

Plugins (1)
  GitHub

Tools (4)
  GitHub issue search         3 calls
  OpenAI Docs search          2 calls
  Shell                       2 calls
  View image                  1 call

Useful interactions could include:

• expand/collapse each category
• show first and most recent use
• show call counts
• distinguish explicit selection from automatic activation
• select an item to jump to the corresponding activity in the conversation
• preserve the inventory when the task is reopened

Privacy and performance requirements

This should be generated from host/runtime activity metadata rather than by asking the model to summarize its own transcript.

By default, retain only:

• task/thread identifier
• Skill, Plugin, or Tool identifier and display name
• timestamps
• invocation count
• explicit versus automatic activation

Do not require storing or displaying prompts, tool arguments, tool outputs, credentials, file contents, or connected-service data.

The inventory should not be injected into the model context and should not consume context-window tokens. Updates should be incremental or lazily rendered so tool-heavy tasks are not slowed down.

Expected behavior

• Each task has its own independent inventory.
• Only capabilities actually used appear.
• Plugin-provided Skills and Tools are attributed to their originating Plugin where known.
• Repeated calls are aggregated rather than duplicated.
• The inventory works for project and projectless tasks.
• The feature remains accurate after task resume, compaction, and app restart.
• If provenance is ambiguous, the UI says so instead of guessing.

Environment

• Surface: Codex in the ChatGPT desktop app
• Platform: Windows
• Request date: 2026-07-16