help team to use ai todo the tedious work:
1. image grouping
2. selecting best shot
3. matching with gps location based on time
4. matching shot to data
and entire harness to manage them, from previewing to manually validate, fix if needed. and custom prompting
opensourcing https://t.co/fM1ABAoTLA #opencode alternative in golang
- advisor support
- automatic permission
- low memory usage, less than 50MB
- able to use chatgpt subscription, deepseek api, xiaomi coding plan, opencode go & zend api
- file explorer
- git explorer
- logs
for all open weights, we can see how capable are these models are. the lesser usage, normally means its higher agentic and coding capabilities. correct me if im wrong
@caps_raunak scary part is when u use the wrong model like deepseek v4 flash, it tend to make mistake and git revert those files silently if you put it on auto mode
@VictorTaelin ive added rules in my agents.md to specifically ask the agent to write to todo.md for anything thats deferred or not worked on for current session, it works wonderfully
i just realised that #claude opus and sonnet code review will give generic non repo based review that does not adhere to agent rules including claude.md, but gpt 5.4 does a better job of understanding the repo and gives less false positive result