#but how was open claw hooking into it?

1 messages ยท Page 1 of 1 (latest)

gaunt shell
#

openclaw was basically just using the 'claude clode' instance that was already installed on your machine. anthropic shut that down by banning any requests that had 'openclaw' in the prompt

rapid scaffold
#

but how did they inject themselves in it. what part of the code.

gaunt shell
#

if you want to dig into the code, just pull down the openclaw repo and ask your agent (codex, claude, whatever) to help you understand it. way more efficient

rapid scaffold
#

no yeah of course. ive had gemini slurp up the code and docs already

#

im just interested in the part where it uses claude code desktop cli in automated fashion

#

but yeah ill ask the ai

gaunt shell
#

the thing with openclaw is, at least until recently, it's always using the pi harness for its llm calls. so, the anthropic calls before we orchestrated by pi, and my understanding is that those were direct calls to anthropic's llm servers and bypassed claude code entirely

more recently, since that's obviously now a very bannable behavior, openclaw is using the 'claude -p <prompt goes here>' cli. You can of course pass in all the extra system prompt stuff, session info, etc. too