[Upstream PR #94] Add anthropic-compatible provider type and custom headers

[quangdang46]

Mirrored from upstream 1jehuang/jcode — Pull Request #94 by @namphamdev
Original state: open
Created: 2026-05-02T02:40:13Z · Updated: 2026-05-02T02:41:32Z
Diff: https://github.com/1jehuang/jcode/pull/94.diff
This issue is an auto-mirrored copy. Comments and edits here are local to quangdang46/jcode — do not expect them to propagate upstream.

---

Summary

• Adds a new \NamedProviderType::AnthropicCompatible\ variant (TOML aliases anthropic-compatible\ / anthropic_compatible) so users can register named profiles that target Anthropic Messages-API gateways.
• Adds a per-profile \headers\ map (\BTreeMap<String, String>) on \NamedProviderConfig\ and wires it through the OpenRouter/OpenAI-compatible runtime so the configured headers are attached to every chat-completion POST.
• Extends \jcode provider add\ with two new flags:
  • --type <openai-compatible|anthropic-compatible>\
  • --header KEY=VALUE\ (repeatable, also accepts \KEY:VALUE)
• Validates header names (ASCII letters/digits/-/_), rejects CR/LF in values, and refuses reserved hop-by-hop names at runtime.
• Persists the chosen type and [providers..headers]\ table to \config.toml; --overwrite\ cleanup now also recognizes the headers subsection.

Files

• \crates/jcode-config-types/src/lib.rs\ - new enum variant + \headers\ field
• \src/cli/args.rs\ / \src/cli/dispatch.rs\ / \src/cli/commands/provider_setup.rs\ - new flags, parsing, TOML emission, tests
• \src/provider/openrouter.rs\ - \parse_named_provider_extra_headers\ + reserved-name guard, struct field, ctor wiring
• \src/provider/openrouter_provider_impl.rs\ / \openrouter_sse_stream.rs\ - thread \extra_headers\ into the request path
• \src/provider/openrouter_tests.rs\ - update test fixtures for the new field

Tests

• \parse_custom_header_args_accepts_kv_and_rejects_invalid\
• \provider_add_writes_anthropic_compatible_type_with_custom_headers\
• \cargo check\ and \cargo check --tests\ are clean for these changes (unrelated pre-existing
  easoning_effort\ errors in \src/overnight.rs\ exist on master too).

Example

\
$ jcode provider add my-claude-gw \
--type anthropic-compatible \
--base-url https://gateway.example.com/v1 \
--model claude-sonnet-4 \
--header anthropic-beta=beta1,beta2 \
--header X-Org=acme \
--api-key-stdin
\\

---

^{Need help on this PR? Tag @codesmith with what you need.}

• Let Codesmith autofix CI failures and bot reviews

[quangdang46]

Deferred. Upstream PR #94 adds a generic "anthropic-compatible" provider type with custom-headers escape hatch. Most of the fork's existing OAI-compat profiles already cover that ground (Cohere, MiniMax, Z.AI, OpenAI-Compatible custom). For users wanting raw-Anthropic-compatibility specifically, they can use jcode provider add with a custom base URL today.

Re-open if you have a specific target endpoint and the existing custom profile flow doesn't fit it.

[quangdang46]

Reopened per user request — closing as 'not planned' was inconsistent with the project's policy of only closing issues that are actually fixed in master. Keeping this open as backlog; the deferral reasoning in the previous comment still applies (needs reproducer / specific environment / upstream port discussion), but the issue stays visible until someone picks it up.