Repository navigation
fix(agents): specialist answers get a configurable 15-minute limit; GPT-profile roles on claude run as a Claude fallback - #338
Merged
Conversation
…PT-profile roles on claude run as a Claude fallback The Workflow pipeline's live run on the lab never completed: every Coding and Testing answer was "stopped by the server" at exactly 5:00. LlmRequest.AgentAnswerTimeout was a hard-coded 5 minutes (since #258), and the roles' gpt-4.1 / gpt-4o-mini models were not passed to Claude Code, so the CLI's default — a slow thinking model on that machine — took 3-5+ minutes per answer. - Agents:AnswerTimeoutMinutes (Agent profiles, default 15, 1-60) replaces the constant, read per call through IAgentProfiles for the Workflow specialists, the Autopilot lead, the ticket decomposer and the retrospective. - The runner maps a non-Claude profile model on claude to REMOTE_AGENT_CLAUDE_FALLBACK_MODEL (default "sonnet"; empty keeps the CLI default) instead of silently running the machine's default model. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018WAdQ7Rm24xss5QDSvXtJB
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
WorkflowAppRealAuthTests(live 5-agent pipeline) never completed — every Coding and Testing answer on the paired runner was "stopped by the server" at exactly 5:00. Root cause (read-only investigation, runner log + code):LlmRequest.AgentAnswerTimeoutwas a hard-coded 5 minutes (since fix(runner): answers never queue behind coding runs; a failed ticket split says so #258), and the roles'gpt-4.1/gpt-4o-minimodels aren't passed to Claude Code, so the CLI's default — a slow thinking model on that machine — ran 3–5+ min per answer. Not a regression from the recent merges.Agents:AnswerTimeoutMinutes(Answer time limit, Settings › Agent profiles; default 15, range 1–60; ADR-0011) replaces the constant, read per call throughIAgentProfilesfor the Workflow specialists, the Autopilot lead, the ticket decomposer and the retrospective. Board runs keep their ownAutopilot:*RunMinuteslimits.clauderuns asREMOTE_AGENT_CLAUDE_FALLBACK_MODEL(defaultsonnet; empty keeps the CLI default; never a flag-like value; operator--modelargs still win) instead of silently becoming the machine's default model.Test plan
AgentProfilesTests(new: default 15 min, tenant value applied per call, no cross-tenant leak, above-Max rejected) +CliModelTests(new: GPT models → fallback, custom fallback, empty fallback, flag-like fallback refused, operator args win) + settings/decomposer/retrospective/orchestrator classes — 217 passedlab.ps1 test→WorkflowAppRealAuthTestscompletes🤖 Generated with Claude Code
https://claude.ai/code/session_018WAdQ7Rm24xss5QDSvXtJB