fix: custom workflow now passes the real task to workers (#87)

[konard]

🤖 Fix: custom workflow now actually builds the project

Fixes #87

Problem

When a user builds their own workflow on the canvas (Boss → Managers → Workers) and submits a task, Octra ran every agent (the chat showed Coordinator done, Developer done, Designer done…) but produced no code — it "started creating the project and immediately stopped, refusing to write code."

Root cause

decisionFromPredefinedWorkflow (orchestrator/internal/service/rules/boss/plan.go) set the TechnicalDescription to the architecture label sent by the frontend:

"Custom workflow with Architect and 2 managers"

That TechnicalDescription is exactly what becomes each worker's TASK.md (boss → manager → worker, via AssignWorkersRequest.TaskMd). So every worker was asked to implement the meaningless label instead of the user's actual request — and generated nothing. The real task (title + description) was silently discarded on the custom-workflow path, even though the AI-planning path always carries a real technical description.

Fix

• TechnicalDescription now carries the real task (title + description); the architecture label is kept only in ArchitectureNotes for Boss validation/context.
• Tech-stack detection on the custom path: when the workflow omits a tech stack (e.g. the Boss node has none configured) it is now detected from the task via detectTechStack, mirroring the AI-planning path — no more silently defaulting every custom workflow to Go.
• Fixed a pre-existing CI-breaking regression in hasRealCode (introduced in 6f5da02): the >= 3 non-comment lines threshold rejected valid short source such as a minimal main.go (package main + func main() {}) or a one-line Express server. This broke TestContainsSourceCode and left orchestrator CI red on master. It now requires >= 1 real code line while still rejecting empty / comment-only stubs (issue #98 intent preserved).

How to reproduce / verify

1. Build a custom workflow on the canvas (Boss + ≥1 Manager + ≥1 Worker), submit any coding task.
2. Before: agents report done, no files are written.
3. After: workers receive the real task and generate code.

Tests

• Updated TestDecisionFromPredefinedWorkflow — asserts the technical description holds the real task and the label lives in ArchitectureNotes.
• Added TestDecisionFromPredefinedWorkflowCarriesRealTask — regression test: the workflow label must not leak into the task, the real request must survive, and the tech stack must be detected (python) when omitted.
• go build ./..., go vet ./..., and go test ./... are green for the orchestrator (and the other Go modules build/test clean).

[konard]

Working session summary

PR #88 is merged. CI on the PR was green (Go Workspace CI + Orchestrator CI both SUCCESS), and the working tree is clean.

• PR: fix: custom workflow now passes the real task to workers (#87) #88 — merged
• Issue: FIX_PROBLEMS_IN_CUSTOM_WORKFLOW #87
• Commit: be228bb fix: custom workflow now passes the real task to workers (#87)

Nothing more to do on this issue — fix is upstream.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Public pricing estimate: $3.594567
• Calculated by Anthropic: $0.555394
• Difference: $-3.039173 (-84.55%)

📊 Context and tokens usage:

Claude Opus 4.8: (2 sub-sessions)

1. 116.0K / 1M (12%) input tokens, 24.1K / 128K (19%) output tokens
2. 48.8K / 1M (5%) input tokens, 805 / 128K (1%) output tokens

Total: (3.7K new + 103.5K cache writes + 3.9M cache reads) input tokens, 24.1K output tokens, $3.222027 cost

Claude Opus 4.7:

• 48.8K / 1M (5%) input tokens, 805 / 128K (1%) output tokens

Total: (13 new + 48.8K cache writes + 95.2K cache reads) input tokens, 805 output tokens, $0.372540 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Main model: Claude Opus 4.8 (claude-opus-4-8)
• Additional models:
  • Claude Opus 4.7 (claude-opus-4-7)

📎 Log file uploaded as Gist (2138KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.