Switch small tier synthesis to per-file cache entries
Add --no-external flag and per-session budget tracking
Add fetch_url tool for URLs referenced in local files
Add ask_user tool and --interactive flag
Add package_lookup tool (PyPI, npm, crates.io, pkg.go.dev)
Research and decide on web search API backend
Create luminos_lib/search.py with web_search implementation
Save investigation plan to cache for resumed runs
Implement dynamic turn allocation from plan output
Implement planning pass with submit_plan tool in ai.py
Add _PLANNING_SYSTEM_PROMPT to prompts.py
Skip survey pass for targets below minimum size threshold
Implement _run_survey() and submit_survey tool in ai.py
Wire survey output into dir loop system prompt
Add _SURVEY_SYSTEM_PROMPT to prompts.py
Add low_confidence_entries() query to CacheManager
Add confidence fields to file and dir cache schemas
Update dir loop prompt to set confidence on cache writes