Feature: Per-workflow log_dir override (Config/Workflow/CLI)

[maddes8cht]

Context

Issue #34 recently established a clean separation between Config (openloop.json) and Workflow JSON fields. The log_dir field currently lives only in Config — it cannot be overridden per workflow or via CLI.

The current cheat-sheet from #34:

Config (openloop.json)	Workflow JSON
agents_dir, workflows_dir, opencode_binary, log_dir, default_max_loops, default_timeout	loop_agents, preparation_agents, finalization_agents, end_state_condition, max_loops, workdir, init_script, opencode_defaults, name

Problem

1. No per-workflow override — log_dir is hard-wired to the global config value. A user running different workflows from different projects cannot point logs to different directories without editing openloop.json each time.
2. No CLI flag — There is --log-file (full path to a single file), but no --log-dir (directory only, letting the engine generate the filename from the workflow name and timestamp). These serve different use cases.
3. Undocumented in schema — log_dir and no_log_file exist in Config (defaults ".openloop" and false) but are missing from the field table in docs/configuration.md.

Proposed Solution

1. Data Model — Config (openloop.json)

Already present: Config.log_dir: str = ".openloop". No change needed.

2. Data Model — Workflow JSON

Add optional log_dir field to the workflow schema:

{
  "name": "my-workflow",
  "log_dir": "./logs/my-project",
  "loop_agents": ["amala"],
  ...
}

• Optional (null/absent = use Config value)
• If relative, resolved against the effective workdir (same behaviour as the current Config-based log_dir)
• Added to WorkflowConfig dataclass (from_dict / to_dict)

3. CLI

New flag --log-dir <path>:

• Highest priority (overrides both workflow and config)
• Accepts relative or absolute paths
• Documented in docs/cli.md

4. Engine (core/engine.py:_init_log())

Override chain (highest to lowest):

CLI --log-dir          -- highest
Workflow JSON log_dir
Config (openloop.json) log_dir   -- lowest

This mirrors the existing pattern for workdir and init_script.

5. GUI (ui/app.py)

The existing Log Directory field in Settings > Environment already exists. It currently reads/writes the Config value. That stays — it acts as the global default. Additionally, the workflow JSON should be able to carry its own log_dir, loaded/saved via _get_workflow_data() / _load_workflow_into_ui().

When a workflow is loaded, its log_dir fills the field (overriding the Config default temporarily). The field always saves back to both: the Config for the next fresh start, and the workflow data if the user saves the workflow.

Implementation options:

• Option A (simpler): The field always writes to both cfg.log_dir and data["log_dir"] on save. The engine's override chain ensures correctness.
• Option B (more explicit): Add a separate "Workflow log dir" field. Probably over-engineered.

6. Documentation

• Add log_dir and no_log_file to the field table in docs/configuration.md
• Add log_dir to the field table in docs/workflows.md
• Add --log-dir to docs/cli.md

Not in Scope

• no_log_file per workflow is intentionally excluded — it's a system-level preference. If a use case arises later it can be added.
• The existing --log-file CLI flag remains as-is (direct file path override).

Implementation Plan

• Add log_dir: Optional[str] = None to WorkflowConfig (core/engine.py)
• Add log_dir serialization to WorkflowConfig.from_dict() / to_dict()
• Add CLI flag --log-dir (openloop.py)
• Update override chain in execute_workflow_data() before calling _init_log()
• Update _init_log() to accept explicit log_dir
• GUI: load workflow log_dir into field, save back to both config and workflow data
• Update docs/configuration.md, docs/workflows.md, docs/cli.md
• Tests: ensure existing 137 unit tests remain green, add new tests for override chain