v0.0.310

Agentic provider receipts, language validity gate, and several CLI fixes.

Features

• The agentic layer now returns structured receipts for provider attempts, covering Codex and opencode evidence (#2425).
• A language validity gate is added to the code generator: generated Python is syntax-checked before being accepted, with a LanguageMismatchError on failure.
• pdd-change Step 6 now correctly emits <pdd-dependency> tags instead of <include> tags (#2376).

Fixes

• ChatGPT subscription auth is now routed through the Responses API, fixing hangs for ChatGPT-backed models (#2316).
• Zsh completion now handles global options (e.g. --model, --env) correctly (#2378).
• Story prompt metadata paths for nested prompts are now resolved correctly (#2380).
• Consumed story rotation replacements are now recognized exactly, preventing false-positive sync drift (#2405).
• Agentic provider evidence handling is hardened: false positives no longer discard valid work, invalid envelopes are rejected, and legacy Codex message shapes are accepted (#2425).

Internal

• Conformance gates (ArchitectureConformanceError, PublicSurfaceRegressionError, TestChurnError, ProseOutputError, LanguageMismatchError) are extracted from code_generator_main into the new pdd/conformance/ package; existing import paths continue to resolve (#2374).
• The cloud gate now runs continuously against main on a 4-hour schedule; make release checks that the candidate SHA has a recent green run before proceeding (#2328, #2329).
• The release branch check is relaxed to require HEAD be an ancestor of origin/main rather than exactly equal to the tip, unblocking releases when the gate blesses a commit that is no longer the tip.

---

What's Changed

• chore(sync): preauthorize cloud-gate paths by @gltanaka in #2329
• fix(release): run the cloud gate continuously instead of during the release by @gltanaka in #2328
• docs(pr-loop): risk-tier review depth and re-review only the delta by @gltanaka in #2330
• fix(release): opt v0.0.310 out of video and repair gate-caught test bugs by @gltanaka in #2332
• fix(sync): bind legacy llm retirement by @gltanaka in #2355
• test(sync): accept schema-3 retirement diagnostics by @gltanaka in #2356
• test(sync): support installed pr2316 authority by @gltanaka in #2359
• chore(sync): install pr2316 transition authority by @gltanaka in #2357
• fix(sync): bind exact pr2316 phase-b states by @gltanaka in #2360
• test(sync): preserve pr2316 phase-b lifecycle states by @gltanaka in #2361
• Fix ChatGPT subscription generation routing by @gltanaka in #2316
• docs: consolidate prompting guide by @DianaTao in #2363
• chore(sync): preauthorize code_generator_main_python language-validity-gate rotation by @agarwal-ishaan in #2382
• Add Language Validity Gate to code_generator_main by @agarwal-ishaan in #2370
• fix: support global options in Zsh completion by @agarwal-ishaan in #2378
• Split code_generator_main into pdd/conformance/ (six gate units) by @agarwal-ishaan in #2374
• fix: pdd-change Step 6 should use tags, not tags by @agarwal-ishaan in #2376
• fix(sync): recognize exact consumed story rotation replacement by @gltanaka in #2405
• test(sync): pin story phase a protected fixture by @gltanaka in #2406
• test(sync): cache immutable story rollout blobs by @gltanaka in #2408
• chore(sync): preauthorize story prompt metadata paths by @gltanaka in #2407
• Fix story prompt metadata paths for nested prompts by @agarwal-ishaan in #2380
• edit prompting guide: frontend prompt and SVG units by @niti-go in #2368
• docs(research): publish Qwen3.8 oQ8e benchmark by @gltanaka in #2409
• chore(sync): preauthorize oMLX Qwen3.8 benchmark research paths by @gltanaka in #2411
• chore(sync): grant absent-path authority for oMLX Pi/Prime benchmark paths by @gltanaka in #2412
• research: benchmark Pi versus Prime on Qwen3.8 by @gltanaka in #2410
• chore(sync): register Pi DSH benchmark paths by @gltanaka in #2415
• chore(sync): preauthorize Pi DSH benchmark paths by @gltanaka in #2416
• research: benchmark Pi versus DeepSeek Harness by @gltanaka in #2414
• research: front-load benchmark caveats and disclosure in Pi-vs-DSH README by @gltanaka in #2417
• prompt: define trustworthy provider-attempt evidence contract by @agarwal-ishaan in #2423
• feat(agentic): return trustworthy provider attempt receipts by @agarwal-ishaan in #2425

New Contributors

• @agarwal-ishaan made their first contribution in #2382

Full Changelog: v0.0.309...v0.0.310