Claude Code for bug bounty
Claude Code can run on any endpoint that speaks Anthropic's Messages API, so you can point it at Kimi K3.1 Pro for notes and report drafts. Its LLM gateway docs list the variables to set.
Setup
export ANTHROPIC_BASE_URL=https://api.silkcompute.comexport ANTHROPIC_AUTH_TOKEN="$SILK_API_KEY"export ANTHROPIC_MODEL=kimi-k3.1-pro-uncensoredexport ANTHROPIC_DEFAULT_HAIKU_MODEL=kimi-k3.1-pro-uncensoredclaudeSource: Claude Code docs. Read November 18, 2026.
Steps
- Set the base URL without /v1
Claude Code adds the /v1/messages path itself.
- Put the key in ANTHROPIC_AUTH_TOKEN
It is sent as a Bearer token.
- Set both model variables
Background tasks use the second one.
- If you get 400 "Extra inputs"
Set CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1.
Use it in a hunt
| Job | Ask |
|---|---|
| Draft a report | Read notes/finding-3.md and draft reports/finding-3.md in HackerOne's sections. Mark gaps with [CHECK]. |
| Check scope | Compare every asset in findings/ with scope.md and list any that match no scope line. |
| Clean your own output | Write a script that merges the hostname files in recon/ and removes duplicates. |
Provider guide: Silk Compute Claude Code setup.