Overview - Ai

ai

7 days30 days

Latest CI Pipeline Executions

Fuzzy

Succeeded
feat/harness-structured-output
c1a9d3aa docs: use latest claude-opus-4-8 in claude-code structured-output example
by Alem Tuzlak
Succeeded
feat/harness-structured-output
c1a9d3aa Merge 9b8778b8812b6c42f3a5e790a9d05d279be5ce5c into 5fcaf90dc82bc20b8c7a75faa3c129da04858af5
by Alem Tuzlak
Succeeded
feat/harness-structured-output
ebf72a46 Merge ccbfc020e0956fac330683f7f92b4c367c6d2c58 into 5fcaf90dc82bc20b8c7a75faa3c129da04858af5
by Alem Tuzlak
Succeeded
feat/harness-structured-output
ebf72a46 feat: honor outputSchema in CLI harness adapters (codex, claude-code, opencode) `chat({ outputSchema })` now works with the Codex, Claude Code, and OpenCode harness adapters. Each declares `supportsCombinedToolsAndSchema() => true` and forwards the JSON Schema to the CLI's native mechanism within the single harness run, so the engine harvests the schema-constrained answer with no separate finalization round-trip: - codex: `codex exec --output-schema <file>`; the final agent_message is buffered and flushed as the sole terminal text (intermediate schema-forced progress messages are dropped, per openai/codex#19816). - claude-code: `claude -p --json-schema '<inline>'`; the result's `structured_output` is emitted as the terminal text and prose is suppressed. - opencode: the session's `json_schema` output format; `message.structured` is emitted as the terminal text and streamed prose is suppressed. The Codex harness throws when `tools` and `outputSchema` are combined, because Codex silently drops the schema when MCP/tools are active (openai/codex#15451). Grok Build now rejects `outputSchema` with an accurate message (its CLI has no schema mechanism); ACP already did. Adds unit tests (translate suppression + command wiring via fake CLIs in real sandboxes) and a gated Claude Code live-smoke case. Updates the codex/claude-code/ opencode adapter docs, the structured-outputs overview, and the adapter-configuration/structured-outputs skills.
by Alem Tuzlak