v0.9.10

Added

• Bedrock guardrails are documented and pinned by a wire contract test. provider_options.bedrock.guardrailConfig reaches the top level of the Converse command through a seam @ai-sdk/amazon-bedrock does not document: its provider-options schema omits the key, and the request builder rest-spreads unknown keys into the command body, so a peer upgrade that tightens that spread would drop guardrails silently. bedrock_guardrail_wire.test.ts drives the real peer over a stubbed fetch and asserts the key lands top-level, survives the effort-translation merge, and that a guardrail_intervened stop surfaces as finish_reason: 'content_filter'. The bedrock section of docs/providers.md gains the matching recipe. No runtime changes.

v0.9.9

No library behavior changes in this release. Every src/ edit in this range is
a documentation comment, so the published runtime is unchanged from v0.9.8; the
work is in the docs surface, the example apps, and the check suite.

Changed

• The public docs surface is now written for the public. research/ and the internal build trees are untracked, and the adoption and comparison pages are back in sync with the native transports: the AI SDK is the default path for seven of the eight providers rather than the only path, and both pages' AI-SDK-backed versus native counts are corrected.
• Every example app now matches docs/blueprint.md. pr-improve, the canonical reference, moved its four system prompts into markdown files with frontmatter, renamed its schemas to snake_case, and moved the field-level output rules out of prose and into .describe() on the schemas, where they travel to the model as part of the JSON schema. red-green-refactor, swebench, and amplify predated the blueprint and were translated: one create_engine call site each, markdown prompts, and state/stages/services splits. amplify's round loop was a while over mutable parent state driving an ensemble whose score callback populated a captured map; it is now loop + map + branch over immutable carry-state, with fallback replacing a buried try/catch, so the stop rule, the fan-out, and the accept/reject decision are all visible in the trajectory.

Fixed

• docs/blueprint.md's state.ts snippet did not compile. It declared its scope-state readers over ReadonlyMap and called .get(), but use(keys, fn) hands fn a plain object projection of the requested keys. The snippet now shows the real shape, with the ReadonlyMap variant documented for readers that go through ctx.state inside a step body instead.
• examples/pr-improve could not run on its own default provider. It defaulted every role to 'sonnet'/'opus' and claimed in comments that the engine expands those per provider. It does not — model is passed to the provider verbatim — so the documented default path (FASCICLE_PROVIDER=anthropic) sent a literal "sonnet" and got a not-found error. There is now one per-provider model table, resolved once and threaded as data, which also removes a second copy of that table and makes the FASCICLE_MODEL_* overrides reach the flow; they were previously parsed and discarded.

Internal

• The check suite is now checkride 0.6.0, replacing the hand-rolled orchestrator. scripts/check.mjs, check-links.mjs, check-doc-snippets.mjs, and check-publish.mjs are deleted (~960 lines) in favour of built-in slots; scripts/check-deps.mjs stays as the one custom check, since the core/engine dependency invariant is fascicle-specific. The pnpm check / check:all / check:json / check:bail contract is unchanged and pnpm check --changed is new.
• Existing debt is grandfathered in checkride.baseline.json (308 diagnostics across lint, struct, dead, dupes, health, spell). The gate ratchets: frozen debt passes, new findings fail. The dupes and health slots are now enforced rather than folded into a single fallow invocation that could not fail on them.
• The packaging gate gains publint and splits the old monolithic publish step into pack, smoke, and attw (pinned to the esm-only profile). Doc snippets are typechecked against the built .d.ts as part of pnpm check:publish, rather than as a separate CI step.
• The blueprint is machine-checked on the examples. All six composition apps carry its ast-grep rules, run for every app by a new examples check slot; check:rules was previously invoked by no CI job and no slot, and the library's own rules never matched the example apps because their files: globs are anchored at the repo root. vitest now globs examples/*/src/**/__tests__, so swebench's and amplify's suites, which had never executed, run in CI: 13 example tests became 51.
• Every module-level function in src/ now carries a doc block, enforced by rules/function-comment-required.yml, and the comments are written in the present tense with no design-history references.
• Tooling: fallow 3.6.0, publint added, and the marketing site dropped from analysis (it reported every file as unused and its DOM handlers as complexity findings).

v0.9.8

Added

• define_agent call-shaping config. DefineAgentConfig accepts model and schema_repair_attempts, closing the gap where the blueprint's "models are threaded as data" rule could not be followed on the fascicle/agents surface (and where schema repair — set on every model call observed in production consumers — required dropping to a stage factory). Precedence: config.model > frontmatter model > engine default; the blueprint's stage-factory section now states the same rule.
• Two blueprint reference apps translated from production consumers, workspace-wired with stub-engine tests and the blueprint's ast-grep rules: examples/change-triage/ (deterministic detectors + one schema-mode model call, a score floor the model cannot undercut, a privacy screen on the model's view) and examples/docs-concierge/ (grounded Q&A with number-based citations, a define_agent stage, and a one-way gate that prefers abstaining over confidently wrong).

Changed

• docs/blueprint.md and site/blueprint.html now state model precedence consistently (threaded model wins, frontmatter is the role default, engine default is the last resort), and the stage-factory snippet no longer implies frontmatter overrides a threaded model.

v0.9.6

Added

• docs/blueprint.md: a standard app architecture for building on fascicle — one composition layer, three adoption tiers, markdown prompts, normalized module shapes, stub-engine testing, and field-observed anti-patterns distilled from the example apps and production consumers.
• Three ast-grep boundary rules wired into examples/pr-improve (runnable via check:rules): create_engine confined to engine.ts, no imperative loops in flow.ts, and fascicle value imports confined to the files allowed to know about it.

v0.9.5

Fixed

• claude_cli provider structured output. compile_schema now strips the top-level $schema (draft-2020-12 URI) and $id keys that z.toJSONSchema (zod v4) stamps, which claude --json-schema rejects. Any model_call on the claude_cli provider that passed a zod schema previously errored out of the box; structured output now works end-to-end.

v0.9.4

Internal

• Hardened mutation-test coverage for the shared engine orchestration core: generate.ts (77.6% -> 95.79%) and tool_loop.ts (83.7% -> 94.59%), via new test suites for streaming dispatch, HITL approval, schema repair, cost/finish accounting, and the timeout/retry machinery.
• Ratcheted the Stryker break/low thresholds from 83 to 84 (full-repo score now 88.33%); documented the remaining equivalent/unreachable mutants inline with Stryker disable annotations rather than adding runtime guards.

v0.9.2

Added

• Native provider transport. Model calls now flow through a neutral native provider contract instead of routing exclusively through the AI SDK. This release lands native Anthropic (mapping, streaming, capabilities, auth), a native OpenAI-compatible core that backs openai, openrouter, and lmstudio, and native Ollama on its /api/chat endpoint. The AI SDK becomes one transport behind the seam rather than the only path, and provider_options pass straight through to the native transport.
• Open provider registry. custom_providers on EngineConfig lets callers register their own providers, and the native provider contract is exported from the package barrel so those providers can be built against a stable type.
• Loop knobs. Per-turn timeout budgets and a prepare_step hook give callers control over each step of the tool loop.
• OpenTelemetry surface. A trajectory-to-otel bridge plus AI SDK telemetry passthrough emit spans for runs and model calls, with the agent-layer boundary ADR amended to record the seam.

Changed

• Breaking: the subprocess provider is renamed to external. Update any provider: "subprocess" configuration to provider: "external".
• The native provider seam is now the default dispatch path; the AI SDK transport is selected behind it rather than assumed.

Internal

• Mutation coverage hardened across the native providers (openai_compatible_native, ollama_native), the otel surface, and with_providers in create_engine, with the Stryker break gate ratcheted up at the final gate.
• Transport parity golden tests plus an OpenRouter live smoke lock the native and AI SDK transports to identical observable behavior.
• The release skill now detects the previous release from the most recent git tag rather than the last vX.Y.Z commit message.

v0.9.1

Internal

• Re-pinned every GitHub Actions dependency to its Node 24 runtime major (actions/checkout v7, actions/setup-node v6, pnpm/action-setup v6, actions/cache v6, actions/deploy-pages v5), clearing GitHub's Node 20 runtime deprecation warning that surfaced on the v0.9.0 publish run. Workflow-only change; the published package is byte-identical to v0.9.0.

v0.9.0

Changed

• Breaking (peer dependencies): AI SDK v7. The ai peer floor is now ^7.0.0 with no v6 compatibility window, and every AI SDK provider peer moves to its v7-line major: @ai-sdk/anthropic ^4, @ai-sdk/openai ^4, @ai-sdk/google ^4, @ai-sdk/amazon-bedrock ^5, @ai-sdk/openai-compatible ^3, @openrouter/ai-sdk-provider ^3, ai-sdk-ollama ^4. Upgrade the whole set together. Fascicle's own API is unchanged: the v7 renames (system to instructions, fullStream to stream, experimental_output to output, stepCountIs to isStepCount) land at the SDK boundary only, so generate({ system }) and friends keep their names.
• Usage and cost accounting is verified against v7's nested token-detail shape (inputTokenDetails/outputTokenDetails, cache-inclusive inputTokens, reasoning-inclusive outputTokens) with concrete-value tests per provider. On @ai-sdk/anthropic v4, cache read and write tokens now fold into inputTokens upstream, which makes fascicle's cost subtraction formula exact instead of skewed for cache-heavy Anthropic calls.

Fixed

• Streamed step_finish chunks now carry the same cached-input/cache-write/reasoning usage granularity as the non-streamed step record (default_usage_from_sdk delegates to the provider usage normalizer).
• The lmstudio (OpenAI-compatible) provider opts into streaming usage (stream_options.include_usage), so streamed calls report real token counts instead of zeros on spec-strict OpenAI-compatible servers such as Ollama's /v1 endpoint.

Internal

• Releases are now published from CI with npm Trusted Publishing (OIDC) and a signed provenance attestation, gated by a required-reviewer GitHub Environment (.github/workflows/publish.yaml). No long-lived npm token exists anywhere; verify a release with npm audit signatures. This makes v0.9.0 the first provenance-attested fascicle publish (see research/provenance-publish-spec.md).
• Agent-layer boundary ADR: fascicle stays on the single-turn seam (generateText/streamText) and declines the v7 agent layer (ToolLoopAgent, WorkflowAgent, HarnessAgent, toolApproval, @ai-sdk/otel), recorded in research/explorations/2026-07-ai-sdk-agent-layer-boundary.md and linked from docs/providers.md.
• examples/live_smoke.ts: a manual release smoke gate that runs one tool-loop flow streamed and non-streamed against OpenRouter and an OpenAI-compatible backend, checking the tool round trip, stream/record text parity, and usage/cost accounting. Live network, so it stays out of the test suite by design.
• Dev toolchain sweep: vitest 4.1.10, oxlint 1.73, oxlint-tsgolint 0.24, tsdown 0.22.3, zod 4.4.3, @types/node 26, fallow 3, and friends.

v0.8.16

Added

• fascicle/ui subpath for streaming a run into an AI SDK useChat UI: to_ui_message_response, pipe_ui_message_stream_to_response, and to_ui_message_chunks translate trajectory model_chunk events onto the AI SDK UI message wire. Ships with docs/human-in-the-loop.md and examples/hitl_http.ts covering suspend/confirm/resume over HTTP.

Changed

• model_call now records model_chunk on the trajectory (via trajectory.record) instead of routing it through ctx.emit, so consumers see a clean top-level model_chunk event carrying the StreamChunk rather than one wrapped in a generic emit.

Fixed

• Disabled the AI SDK's built-in retry (maxRetries: 0) on the generate and stream calls. The engine already owns retry via its retry policy; the SDK's default of 2 nested a second retry loop inside every attempt, inflating provider round-trips and distorting backoff.