Fix field type inference for target_repo and worker_workflow in update_project

[Copilot]

The update_project safe output handler incorrectly inferred target_repo and worker_workflow as SINGLE_SELECT fields when they should be TEXT fields, causing repeated warnings during project board updates.

Changes

• Field type detection logic (update_project.cjs)
  • Added target_repo and worker_workflow to TEXT field classification
  • Updated 4 instances of field type inference (lines 666, 701, 845, 880)

Context

The field type inference uses heuristics to determine if a field should be TEXT vs SINGLE_SELECT. Fields like campaign_id and classification were already classified as TEXT, but target_repo and worker_workflow were falling through to the default SINGLE_SELECT type.

// Before
const isTextField = "classification" === fieldName.toLowerCase() || 
                    "campaign_id" === fieldName.toLowerCase() || 
                    ("string" == typeof fieldValue && fieldValue.includes("|"));

// After  
const isTextField = "classification" === fieldName.toLowerCase() || 
                    "campaign_id" === fieldName.toLowerCase() || 
                    "target_repo" === fieldName.toLowerCase() || 
                    "worker_workflow" === fieldName.toLowerCase() || 
                    ("string" == typeof fieldValue && fieldValue.includes("|"));

Note

The workflow logs showed "only 10 project update items" - this is expected behavior, not a bug. The agent generated 10 safe-output messages, and all were processed correctly with max: 100 configured.

Original prompt

Analyze workflow run:
https://github.com/githubnext/gh-aw/actions/runs/21437319107/job/61731511388

• Check why only 10 project update items were allowed
• Fix warnings (field types, etc.)

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.