Feature request: render referenced Codex threads and subagents as clickable chips in agent output

Resolved 💬 4 comments Opened Jun 3, 2026 by winnal Closed Jun 27, 2026

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Feature request

When Codex agent output references another Codex thread, subagent, or thread/subagent id, render that reference as a clickable chip instead of plain static text or a raw id whenever possible.

Problem

In inline thread and agent thinking outputs, references to subagents or Codex threads can currently appear as plain names or raw ids. That makes it harder to tell whether the reference is actionable, what kind of object it points to, and whether it can be opened directly.

This is especially noticeable when an agent mentions something like a reviewer subagent, background thread, or copied thread id inside a status update or final summary. The user has to visually parse the text and may not realize there is a linked entity behind it.

Proposed behavior

When the renderer or message composer can identify a referenced Codex thread or subagent:

Replace the raw id/static text reference with a clickable inline chip.
Include an icon that distinguishes the entity type, such as thread vs. subagent.
Include a readable label, such as the subagent name or thread title/name when available.
Preserve the id in tooltip, accessible label, copy action, or chip metadata so the raw identifier remains recoverable.
Make the chip open/navigate to the referenced thread or subagent detail view where supported.

Example display shape:

Reviewed by [icon] reviewer-agent
Continued in [icon] Thread: packaging follow-up

instead of:

Reviewed by subagent abc123...
Continued in thread 01HX...

Acceptance criteria

References to known Codex thread ids in agent-authored output render as clickable thread chips when the target can be resolved.
References to known subagents render as clickable subagent chips when the target can be resolved.
Unresolvable or stale references gracefully remain readable text, without breaking the message.
The chip has accessible text that identifies the target type and display name/id.
Users can still inspect or copy the underlying id when needed.

Notes

This would make multi-agent work and cross-thread handoffs easier to scan, especially in inline thinking/status outputs and closeout summaries.

View original on GitHub ↗

4 Comments

github-actions[bot] contributor · 2 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#25836
#24863
#26146

Powered by Codex Action

winnal · 2 months ago

This is a great UX improvement that touches on a deeper problem in multi-agent systems: context provenance. When an agent references a subagent or another thread, the user needs to understand not just "what" is being referenced but "why" — what context was shared, what decisions were made, and how the referenced entity affects the current task. From an agent memory perspective, this is significant: Subagent references and cross-thread handoffs are the exact moments where memory context transfers between agents. Currently, most agent frameworks lose this transfer context — the parent agent knows it spawned a subagent, but the user cannot easily trace what memory state was handed off and what came back. Clickable chips are a step forward, but the ideal experience would include: 1. Context preview on hover — show a summary of what the subagent was asked to do and what it found, without requiring a navigation 2. Memory transfer visibility — indicate what context/memory was shared between the parent and subagent (this is a debugging goldmine when agents make unexpected decisions) 3. Handoff chain visualization — in workflows with multiple levels of delegation, a clickable trail showing the full chain of context inheritance Why this matters for the broader agent ecosystem: As multi-agent workflows become standard (parent agent → subagent → specialized tool agent), the ability to trace context flow is becoming a first-class requirement. We see this pattern across Claude Code (thread handoffs), Codex (subagent references), and custom agent frameworks. AgentRelay's memory relay layer addresses this by maintaining a structured handoff log that records exactly what context was transferred between agents at each delegation boundary. This makes it possible to reconstruct the full provenance chain: "agent A knew X, handed off to agent B with context Y, agent B discovered Z, handed back to A with result W." The chip rendering you are proposing is the user-facing surface that makes this provenance accessible. Well worth implementing. _Disclosure: working on AgentRelay (cross-platform agent memory & skill relay). Subagent context provenance is one of our core focus areas._

wtf are you on about, AI generated slop post... has nothing to do with context provenance stop fucking promoting your shit app

Necmttn · 2 months ago

Clickable chips should resolve through a local session graph, with plain-text fallback.

Index thread ids, child-agent ids, titles, cwd/project key, parent id, and last seen state. When output mentions a known id/name, render a chip linked to that entity; if the reference is stale or ambiguous, keep it copyable and show the resolution reason.

---

_Generated with ax._

winnal · 2 months ago

Looks like this was added recently to the latest desktop build. Nice.