Triage agent hardcoded categories

[deboer-tim]

The triage agent has hardcoded categories and associated labels today, e.g. bug, documentation, performance, feature, and 'other'. It only triggers the coding agent for the first three as 'low risk categories'.

This behaviour has been confusing for users:

• A project may have more (or less) existing categories, the agent doesn't pick up on it and file things correctly.
• The comment 'looks the same' regardless of category or if it's going to trigger the coding agent or not: you need to look at the labels to see how it decided and whether it triggered the coder or not. You need to understand its logic.
• Bug fixes can be high risk; many features are low risk. The default seems arbitrary and should be based on risk or complexity instead.
• To customize any of this you'd have to start by copying a 580 line script that you couldn't maintain or keep in sync with this repo...

[fullsend-ai-triage]

🤖 Finished Triage · ✅ Success · Started 2:47 PM UTC · Completed 2:50 PM UTC
Commit: 122e680 · View workflow run →

[fullsend-ai-triage]

This issue describes the same root problem as #618 — the triage agent's hardcoded category set causes misclassification, opaque routing, and poor customizability. That issue already has a detailed proposed implementation (the TRIAGE_EXTRA_CATEGORIES env var) and references the same downstream pain points.

The specific concerns raised here are being addressed across several related efforts:

• Categories don't match project needs / customization requires forking → triage: support user-defined categories via TRIAGE_EXTRA_CATEGORIES #618 proposes TRIAGE_EXTRA_CATEGORIES to extend categories via env vars without forking the schema or scripts.
• Auto-dispatch based on category rather than risk/complexity → PR feat(triage): effort-estimation skill and auto-promotion gate #646 introduces an effort-estimation skill that gates auto-promotion on risk/complexity scoring instead of category membership.
• Routing configurability → PR feat(#1754): add TRIAGE_AUTO_CODE configuration for ready-to-code #567 adds TRIAGE_AUTO_CODE to control which categories auto-promote to ready-to-code.

The UX concern about the triage comment not indicating whether the coding agent will be triggered is a valid point — it may be worth filing as a separate, focused issue if it isn't addressed by the work above.

Closing in favor of #618 as the canonical tracking issue. The related PRs (#567, #646) cover the routing and risk-assessment dimensions.

[deboer-tim]

Reopening: #618 addresses the category issue (first and last bullet points), but neither of the middle two bullets (disconnect between comment and action, poor default risk assessment).

Triage agent is still closing an issue when any part is duplicate/solved instead of looking at all points raised.

[ralphbean]

I think what I want here is for triage agent to decompose your issue so that two could be closed as duplicates and two left open (linked, related - possibly as sub-issues of an "epic" type issue).

[fullsend-ai-prioritize]

RICE Priority Score: 0.93

Score breakdown

Dimension	Score	Reasoning
Reach	2	The triage agent is a core workflow component that all users interact with. The confusing comment UX and arbitrary risk-based routing affect most active users across organizations, not just a single project. Filed by a collaborator (deboer-tim) indicating broad relevance.
Impact	1	The two remaining unaddressed concerns—triage comment not indicating whether the coding agent will be triggered and risk assessment based on category rather than complexity—cause noticeable workflow friction but are not blocking. Users can work around them by inspecting labels and manually triggering the coder. This is a medium improvement to workflow clarity and routing accuracy.
Confidence	0.7	The issue is well-described with clear bullet points, and comments clarify which sub-problems remain open (bullets 2 and 3). However, some related work is in-flight (PR #646 for effort estimation, PR #567 for TRIAGE_AUTO_CODE), making the residual scope somewhat uncertain.
Effort	1.5	Two distinct changes are needed: updating the triage comment template to surface routing decisions (relatively straightforward) and reworking auto-dispatch to use risk/complexity scoring instead of category membership (partially addressed by PR #646 but still requires integration and design work). Multiple components are touched, warranting moderate effort.

Formula: (2 x 1 x 0.7) / 1.5 = 0.93