Claude processes started by T3 Code identify themselves with CLAUDE_CODE_ENTRYPOINT. Today that value is the Agent SDK default sdk-ts, or a value inherited from the host. For example, it is claude-desktop when T3 Code is started from inside a Claude Code session.
mergeClaudeProcessEnvironment in ClaudeHome.ts sets CLAUDE_CODE_ENTRYPOINT=t3-code for the Claude processes T3 Code runs: turns, version and capability probes, and text generation.t3-code.settings.json env wins over both. Claude Code 2.1.316 reads the entrypoint only from the process env and ignores the one in settings.json, so T3 Code reads it from there itself. readClaudeSettingsBaseUrl became readClaudeSettingsEnv so that one reader serves both values.claude started there keeps its own entrypoint.ClaudeHome.test.ts covers three cases: the default replaces an inherited host value, the instance env overrides the default, and settings.json overrides both.CLAUDE_CODE_ENTRYPOINT=t3-code in the process env, the request carries User-Agent: claude-cli/2.1.316 (external, t3-code) and cc_entrypoint=t3-code.t3-code in the process env and a different value in settings.json, the request still carries t3-code. This is why T3 Code applies the settings.json value itself.vp run check:changed passes.Made with Claude Opus 5.5 in Claude Code, running in T3 Code.
🤖 Generated with Claude Code
Added 1 commit
Verdict: Needs Human Approval
Macroscope's review found this PR requires human approval — Claude launches now default their entrypoint identifier to "t3-code", while allowing instance and settings overrides. This changes an existing product default and requires human confirmation of the intended identity and override precedence.
You can add or adjust custom eligibility rules. Learn more.
✅ Thread transfer remains within every enforced ceiling.
| Provider | Metric | Main baseline | This PR | Impact | PR ceiling | |
|---|---|---|---|---|---|---|
| Codex | Total thread wire | 5.0 KiB | 5.0 KiB | +35 B (+0.7%) | 6.8 KiB | ✅ |
| Codex | Thread snapshot wire | 3.8 KiB | 3.8 KiB | 0 B (0.0%) | 4.9 KiB | ✅ |
| Codex | Live turn WebSocket wire | 1.2 KiB | 1.3 KiB | +35 B (+2.8%) | 2.0 KiB | ✅ |
| Codex | Live turn WebSocket decoded | 21.5 KiB | 21.6 KiB | +41 B (+0.2%) | 24.4 KiB | ✅ |
| Codex | Live turn messages | 1 | 2 | +1 (+100.0%) | 8 | ✅ |
| Claude | Total thread wire | 5.0 KiB | 5.0 KiB | 0 B (0.0%) | 6.8 KiB | ✅ |
| Claude | Thread snapshot wire | 3.8 KiB | 3.8 KiB | 0 B (0.0%) | 4.9 KiB | ✅ |
| Claude | Live turn WebSocket wire | 1.3 KiB | 1.3 KiB | 0 B (0.0%) | 2.0 KiB | ✅ |
| Claude | Live turn WebSocket decoded | 21.9 KiB | 21.9 KiB | 0 B (0.0%) | 24.4 KiB | ✅ |
| Claude | Live turn messages | 2 | 2 | 0 (0.0%) | 8 | ✅ |
Baseline: 575579d · PR result: 4d75fb8 · Source CI: success
10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.
No actionable comments were generated in the recent review.
77d725fb-c6d4-44d3-819c-fb7e65e0d732Reviewing files that changed from the base of the PR and between 575579d68ac4bae09f8426a345d4a3eac1e90c67 and 4d75fb830717ef5d9b5671534a115141c56e9995.
apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.tsapps/server/src/provider/ClaudeProvider.tsapps/server/src/provider/Drivers/ClaudeDriver.tsapps/server/src/provider/Drivers/ClaudeHome.test.tsapps/server/src/provider/Drivers/ClaudeHome.tsapps/server/src/provider/Drivers/ClaudeSkills.tsIncluded review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.
Claude settings now provide both ANTHROPIC_BASE_URL and CLAUDE_CODE_ENTRYPOINT. Claude process-environment merging applies the entrypoint from settings when it is nonempty. The Claude driver and orchestration adapter use this merger.
Changes: Claude environment resolution
| Layer / File(s) | Summary |
|---|---|
Read and merge Claude settings environment apps/server/src/provider/Drivers/ClaudeSkills.ts, apps/server/src/provider/Drivers/ClaudeHome.ts, apps/server/src/provider/Drivers/ClaudeHome.test.ts | The settings reader returns both environment values, with later settings files overriding earlier values for fields they specify. The merge function combines host and instance environments and applies a nonempty settings entrypoint. Tests cover entrypoint precedence and preservation of a host value. |
Use Claude environment resolution across provider paths apps/server/src/provider/Drivers/ClaudeDriver.ts, apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts, apps/server/src/provider/ClaudeProvider.ts | The driver and orchestration adapter use the Claude-specific merger. The provider reads ANTHROPIC_BASE_URL through the expanded settings reader and retains its process-environment fallback. |
Estimated code review effort: 2 (Simple) | ~10 minutes
Change: Bug fix
Suggested reviewers: maria-rcks
4d75fProject-specific entrypoint overrides may be ignored, affecting attribution. This is a bounded merge risk.
Failed Failed checks: 1 warning
| Check name | Status | Explanation | Resolution |
|---|---|---|---|
| Description check | Warning | The description explains the problem, implementation, behavior, and verification results. However, it omits the required "## Problem" and "## Scope and approval" sections, including an issue or mainta… | Add the required "## Problem" and "## Scope and approval" sections. Include the triaged issue or explicit maintainer approval, or explain why this focused fix qualifies without prior approval. Keep the existing change and verification detai… |
| Check name | Status | Explanation |
|---|---|---|
| Title check | Passed | The title clearly identifies the main change: updating the Claude entrypoint identifier. It is concise and specific. |
| Linked Issues check | Passed | Check skipped because no linked issues were found for this pull request. |
| Out of Scope Changes check | Passed | Check skipped because no linked issues were found for this pull request. |
Explanation
The description explains the problem, implementation, behavior, and verification results. However, it omits the required "## Problem" and "## Scope and approval" sections, including an issue or maintainer approval link or an explanation for why the focused fix qualifies without one.
Resolution
Add the required "## Problem" and "## Scope and approval" sections. Include the triaged issue or explicit maintainer approval, or explain why this focused fix qualifies without prior approval. Keep the existing change and verification details under their required headings.
Astra (AI assistant) writing on behalf of @RCXDd. Adding a measured CLIProxyAPI compatibility check for this entrypoint change.
On November 2, I compared a genuine interactive Claude Code TUI (cli) with the same binary running headless through CLIProxyAPI with CLAUDE_CODE_ENTRYPOINT=t3-code. Both entrypoints were verified in the session transcripts.
| Measurement | Native TUI | Proxy with t3-code |
|---|---|---|
| Total input tokens processed, including cache hits | 48,746 | 48,184 |
| Output tokens | 110 | 124 |
| Final-turn input served from cache | 99.713% | 99.710% |
Native usage was deduplicated by message/request identity from transcripts. Proxy totals came from the final cumulative modelUsage, avoiding partial streamed usage records. This small workload showed essentially equivalent input-token and warm-cache efficiency.
Proxy-specific behavior: CLIProxyAPI 8.0.50's native entrypoint allowlist includes sdk-ts but not t3-code. With our default OAuth configuration, its source routes t3-code through the rewriting path, which emits cli upstream and adjusts system-prompt/cache handling. Thus the new label also changes the proxy's processing path; these tests exercised that path successfully.
Scope: this reproduces the PR's environment-variable change, rather than testing an installed nightly or the full T3 instruction set. It is not proof of equal subscription quota accounting: the tested Ultra account reports five_hour: null, and the whole-percentage weekly meter cannot resolve this small workload. The proxy also began with 1,061 cached prefix tokens, so initial cache-write totals would not be a fair cost comparison. Longer sessions, compaction, and subagents were not tested.
If further quota verification is needed, the missing evidence is a controlled run on an account that still has a five-hour cap. No credentials, account identifiers, private endpoints, or raw session logs are included here.
Honestly it would be awesome to see my name in this PR https://github.com/pingdotgg/t3code/pull/25597#event-34745329984
He's the GOAT. Thank you!
