Fix #5931: Propagate Crew.prompt_file to all components during execution

[devin-ai-integration]

Summary

Fixes #5931

Setting prompt_file on a Crew previously had no effect on agents, tasks, tools, or any other components during execution. The custom I18N instance was only used for the manager agent's role/goal/backstory in hierarchical mode, while all other code paths used the hardcoded I18N_DEFAULT module-level singleton.

Root cause: Every file that needed prompt text imported and used I18N_DEFAULT directly — a module-level constant that always points to the built-in en.json prompts. The Crew.prompt_file parameter was never propagated to the components that generate system prompts, task prompts, tool descriptions, error messages, memory formatting, etc.

Fix: Introduced a contextvars.ContextVar-based mechanism that scopes a custom I18N instance to the crew's execution lifetime:

1. Added get_crew_i18n() / set_crew_i18n() / reset_crew_i18n() to i18n.py — when no override is set, get_crew_i18n() falls back to I18N_DEFAULT, so default behavior is unchanged.
2. prepare_kickoff() sets the context var when prompt_file is provided on the crew.
3. Crew.kickoff() resets the context var in its finally block (even on exceptions).
4. Replaced all I18N_DEFAULT references across 29 source files with get_crew_i18n() calls, so every component (Prompts, Task, agent tools, tool usage, memory, reasoning, planners, etc.) now respects the crew's prompt_file.

This approach is:

• Thread-safe: each crew execution has its own context via contextvars
• Backward-compatible: get_crew_i18n() returns I18N_DEFAULT when no crew override is active
• Minimal surface area: no new fields on Agent/Task, no signature changes

Review & Testing Checklist for Human

• Verify that a custom prompt_file on a Crew actually changes the system/task prompts during execution (create a crew with prompt_file pointing to a custom JSON file, run kickoff, and check that prompts contain the custom text)
• Verify that crews without prompt_file still use default prompts (regression check)
• Verify hierarchical process still works correctly with prompt_file (manager agent + delegation tools should use custom prompts)
• Check that concurrent/nested crews with different prompt_file values don't leak I18N state between each other
• Inspect the get_crew_i18n() replacements in a few key files (prompts.py, task.py, tool_usage.py) to confirm the mechanical replacement is correct

Notes

• telemetry.py was intentionally left using I18N_DEFAULT since it only reports the prompt_file metadata value, not actual prompt content.
• Class-level field defaults using I18N_DEFAULT in default_factory lambdas (e.g., AddImageTool.name) were also updated to use get_crew_i18n() — since these lambdas execute at instantiation time (during crew execution), they correctly pick up the context var.
• The lru_cache on get_i18n() means each unique prompt_file path only triggers one file read/JSON parse across the entire process lifetime.
• 12 new tests were added covering the context variable mechanism, Prompts class, Task.prompt(), AgentTools, and full Crew.kickoff() integration.

Link to Devin session: https://app.devin.ai/sessions/67e13fa57a594ef0ab900d1d73643ba3

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring