Context
This is a follow-up to #64, which was closed with the conclusion that
nanoclaw's use of the Claude Agent SDK via subscription credentials was
a matter of individual risk judgement.
However, the situation has changed significantly since then.
What Changed
Shortly after that discussion, Anthropic explicitly clarified in their
official compliance documentation that using Claude Free/Pro/Max
subscription OAuth tokens with the Agent SDK — or any third-party tool —
is a violation of the Consumer Terms of Service:
OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.
References:
Proposal
Fortunately, nanoclaw's dependency on the Agent SDK appears to be quite
minimal — primarily centered around the query() call. This suggests
that replacing it with a direct claude CLI invocation (e.g., via
child_process or subprocess) could be a relatively contained change.
Claude Code CLI itself is an official Anthropic product, and using it
for local automation and agentic workflows — particularly when
authenticated via ANTHROPIC_API_KEY — appears to be within the
permitted use cases outlined in the documentation.
Questions for the Maintainer
- Is there an existing plan or roadmap item to migrate away from the
Agent SDK toward direct Claude Code CLI invocation?
- Would a community contribution toward this migration be welcome?
Given the updated policy clarity, I think it's worth revisiting this
as a project priority rather than leaving it to individual users to
assess their own risk. Would love to hear the maintainer's thoughts.
Related issue: #64
Context
This is a follow-up to #64, which was closed with the conclusion that
nanoclaw's use of the Claude Agent SDK via subscription credentials was
a matter of individual risk judgement.
However, the situation has changed significantly since then.
What Changed
Shortly after that discussion, Anthropic explicitly clarified in their
official compliance documentation that using Claude Free/Pro/Max
subscription OAuth tokens with the Agent SDK — or any third-party tool —
is a violation of the Consumer Terms of Service:
References:
Proposal
Fortunately, nanoclaw's dependency on the Agent SDK appears to be quite
minimal — primarily centered around the
query()call. This suggeststhat replacing it with a direct
claudeCLI invocation (e.g., viachild_process or subprocess) could be a relatively contained change.
Claude Code CLI itself is an official Anthropic product, and using it
for local automation and agentic workflows — particularly when
authenticated via
ANTHROPIC_API_KEY— appears to be within thepermitted use cases outlined in the documentation.
Questions for the Maintainer
Agent SDK toward direct Claude Code CLI invocation?
Given the updated policy clarity, I think it's worth revisiting this
as a project priority rather than leaving it to individual users to
assess their own risk. Would love to hear the maintainer's thoughts.
Related issue: #64