Implement 0086 PromptManager default cache TTL

CHANGELOG.md

@@ -4,6 +4,12 @@ All notable changes to `openarmature-python` are documented in this file.
 
 The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). The package follows [Semantic Versioning](https://semver.org/); pre-1.0 minor bumps may carry behavioral changes per [spec governance](https://github.com/LunarCommand/openarmature-spec/blob/main/GOVERNANCE.md).
 
+## [0.17.0] - 2026-07-20
+
+### Added
+
+- **PromptManager service-wide default cache TTL** (proposal 0086, prompt-management §6, spec v0.79.0). `PromptManager` construction accepts an optional `default_cache_ttl_seconds`, applied to any `fetch` or `get` that omits a per-call `cache_ttl_seconds`. Resolution follows a precedence chain: an explicit per-call value (including `0` force-fresh) wins; otherwise the manager default applies; otherwise nothing is forwarded and the backend's own caching governs. An omitted or explicit-`None` per-call value both select the default, so resolution does not depend on argument presence. A negative default is rejected at construction, and the per-call negative rejection is unchanged. `get` delegates to `fetch` and inherits the chain, and the bundled caching backends need no change (they already honor a resolved `cache_ttl_seconds`). Conformance fixture 036 is un-deferred.
+
 ## [0.16.0] — 2026-07-18
 
 ### Added

conformance.toml

@@ -840,10 +840,12 @@ since = "0.16.0"
 note = "The SAVE-side enclosing_fan_out_lineage keying (pipeline-utilities §10.11) shipped in #194: a fan-out instance's checkpoint tracking key carries the enclosing fan-out instance lineage in the in-memory dict and through the checkpoint projection / lookup / cleanup / restore, so concurrent outer instances no longer collide. partial because the RESUME consume-side is not yet shipped: a fan-out nested inside an outer instance re-runs rather than skipping on resume, since the saved record format carries no lineage (tracked as a follow-up). pipeline-utilities fixture 076 is not collected by the test_pipeline_utilities.py _LAST_DRIVEN_FIXTURE number gate (it is not deferred); the resume consume-side plus its fixture wiring land in a later PR."
 
 # Spec v0.79.0 (proposal 0086).  Service-wide default cache_ttl_seconds
-# on PromptManager (prompt-management §6).  Not-yet; prompt-management
-# fixture 036 defers with it.
+# on PromptManager (prompt-management §6).  Implemented since 0.17.0;
+# prompt-management fixture 036 pins it.
 [proposals."0086"]
-status = "not-yet"
+status = "implemented"
+since = "0.17.0"
+note = "PromptManager construction gains an optional default_cache_ttl_seconds (prompt-management §6). fetch/get resolve cache_ttl_seconds by precedence: an explicit per-call value (including 0 force-fresh) wins; else the manager default; else None (the backend's own caching governs). An omitted or explicit-None per-call value both select the default, so resolution does not depend on argument-presence semantics. A negative default is rejected at construction; the per-call negative rejection from 0072 is unchanged and applies to the per-call value before resolution. get() delegates to fetch, so it inherits the chain unchanged; the bundled caching backends already honor a resolved cache_ttl_seconds (0072), so no backend change. Fixture 036 (default applied on omit + advance-clock aging + per-call 0 override) is un-deferred and driven via a manager: {default_cache_ttl_seconds} construction slot + target: {manager: true} fetches."
 
 # Spec v0.82.0 (proposal 0087).  Conformance-adapter within-node
 # directive execution order (§8.3).  Implemented since 0.17.0;

docs/concepts/prompts.md

@@ -73,6 +73,32 @@ recent = await manager.get(
 )
 ```
 
+### A service-wide default
+
+If most fetches want the same freshness bound, set it once at construction
+with `default_cache_ttl_seconds` instead of passing `cache_ttl_seconds` on
+every call:
+
+```python
+# `backend` is any backend with a client-side cache (e.g. the Langfuse backend);
+# a cacheless backend ignores the TTL regardless of where it comes from.
+manager = PromptManager(backend, default_cache_ttl_seconds=60)
+
+# Uses the default (60s); no per-call value needed:
+prompt = await manager.fetch("greeting", "production")
+
+# A per-call value always wins, so this still force-refreshes:
+fresh = await manager.fetch("greeting", "production", cache_ttl_seconds=0)
+```
+
+Resolution follows a precedence chain: an explicit per-call value (including
+`0`) wins; otherwise the manager default applies; otherwise nothing is
+forwarded and the backend's own caching governs. A negative default is
+rejected at construction. Once a default is set, an omitted per-call value
+resolves to it, so there is no per-call way to defer to the backend's own
+behavior for a single fetch while a default is configured. Configure no
+default, or pass an explicit value, if you need that.
+
 ## Prompt identity
 
 Every `Prompt` carries five identity fields:

src/openarmature/prompts/manager.py

@@ -76,12 +76,24 @@ def __init__(
         self,
         *backends: PromptBackend,
         label_resolver: LabelResolver | None = None,
+        default_cache_ttl_seconds: int | None = None,
         jinja_undefined: type[jinja2.Undefined] = jinja2.StrictUndefined,
     ) -> None:
         if not backends:
             raise ValueError("PromptManager requires at least one backend")
+        # Proposal 0086 (prompt-management §6): a service-wide default
+        # cache_ttl_seconds, applied to any fetch/get that omits a per-call
+        # value. Same per-fetch semantics as the per-call lever (0 = force
+        # fresh, N > 0 = bound staleness to N seconds); a negative default
+        # is invalid and is rejected at construction.
+        if default_cache_ttl_seconds is not None and default_cache_ttl_seconds < 0:
+            raise ValueError(
+                f"default_cache_ttl_seconds must be >= 0 (got {default_cache_ttl_seconds!r}); "
+                "None leaves per-fetch cache control to the backend, 0 forces fresh reads"
+            )
         self._backends: tuple[PromptBackend, ...] = backends
         self._label_resolver = label_resolver
+        self._default_cache_ttl_seconds = default_cache_ttl_seconds
         # autoescape disabled by design: render output goes to an LLM
         # API call (plain text), not an HTML response. The env is
         # per-manager (was module-level) so jinja_undefined can be
@@ -127,21 +139,30 @@ async def fetch(
           failures, the manager raises ``PromptStoreUnavailable``.
 
         ``cache_ttl_seconds`` is a read-side cache control forwarded to
-        each backend's ``fetch``: ``None`` keeps
-        current behavior, ``0`` forces a fresh read, ``N > 0`` bounds a
-        served entry's staleness to N seconds; a negative value is
-        rejected. Cacheless backends ignore it.
+        each backend's ``fetch``: ``0`` forces a fresh read, ``N > 0``
+        bounds a served entry's staleness to N seconds, and a negative
+        value is rejected. When omitted (or ``None``), the manager's
+        ``default_cache_ttl_seconds`` applies if one was configured;
+        otherwise nothing is forwarded and the backend's own caching
+        governs. An explicit per-call value always overrides the default.
+        Cacheless backends ignore the resolved value.
         """
         if cache_ttl_seconds is not None and cache_ttl_seconds < 0:
             raise ValueError(
                 f"cache_ttl_seconds must be >= 0 (got {cache_ttl_seconds!r}); "
-                "None preserves current behavior, 0 forces a fresh read"
+                "None selects the manager default or the backend's own caching, "
+                "0 forces a fresh read"
             )
+        # Proposal 0086 precedence: an explicit per-call value (including
+        # 0) wins; else the manager default; else None (the backend's own
+        # caching governs). An omitted or explicit-None per-call value both
+        # select the default, so resolution is presence-independent.
+        resolved_ttl = cache_ttl_seconds if cache_ttl_seconds is not None else self._default_cache_ttl_seconds
         resolved_label = self._resolve_label(label, name)
         causes: list[BaseException] = []
         for backend in self._backends:
             try:
-                return await backend.fetch(name, resolved_label, cache_ttl_seconds=cache_ttl_seconds)
+                return await backend.fetch(name, resolved_label, cache_ttl_seconds=resolved_ttl)
             except PromptNotFound:
                 raise
             except PromptStoreUnavailable as exc:

tests/conformance/harness/prompt_management.py

@@ -74,8 +74,12 @@ class FixtureLabelResolverSpec(_StrictModel):
 
 
 class FixtureManagerSpec(_StrictModel):
-    backends: list[str]
+    # backends is optional: fixture 036's manager block omits it and
+    # defaults to all declared backends in order (proposal 0086).
+    backends: list[str] | None = None
     label_resolver_ref: str | None = None
+    # Proposal 0086: the manager's service-wide default cache_ttl_seconds.
+    default_cache_ttl_seconds: int | None = None
 
 
 # ---------------------------------------------------------------------------
@@ -87,8 +91,17 @@ class BackendTarget(_StrictModel):
     backend: str
 
 
+class ManagerTarget(_StrictModel):
+    # Proposal 0086: fixture 036 routes fetches through the manager via the
+    # dict form ``target: {manager: true}`` (mirroring ``{backend: <name>}``),
+    # alongside the bare-string ``manager`` the other fixtures use. Fixed to
+    # ``true`` -- ``{manager: false}`` is nonsensical and rejected at parse.
+    manager: Literal[True]
+
+
 CallTarget = (
     BackendTarget
+    | ManagerTarget
     | Literal[
         "manager",
         "secondary_manager",

tests/conformance/test_fixture_parsing.py

@@ -636,11 +636,6 @@ def _id(case: tuple[str, Path]) -> str:
         "directive shapes the cross-cap parser does not model, "
         "runtime-driven in test_observability"
     ),
-    # Proposal 0086 (PromptManager default cache_ttl_seconds, v0.79.0) -- the
-    # manager default-cache-ttl directive shape.
-    "prompt-management/036-prompt-manager-default-cache-ttl": (
-        "Proposal 0086 default cache_ttl_seconds; not implemented"
-    ),
 }
 
 

tests/conformance/test_prompt_management.py

@@ -43,10 +43,12 @@
 from ._deferral import skip_if_deferred
 from .harness.loader import CONFORMANCE_ROOT
 from .harness.prompt_management import (
+    BackendTarget,
     FixtureBackendSpec,
     FixtureCall,
     FixtureExpectedResultEquivalence,
     FixtureManagerSpec,
+    ManagerTarget,
     PromptManagementFixture,
 )
 
@@ -339,10 +341,13 @@ async def _run_call(
             members = [captures[ref] for ref in call.members_refs]
             return PromptGroup(group_name=call.group_name, members=members), None
 
-        if isinstance(target, str) and target in {"manager", "secondary_manager", "tertiary_manager"}:
-            # All three manager targets dispatch to the currently-active
-            # manager in the per-pair iteration loop. The naming exists
-            # only to keep fixture YAML self-describing under a
+        if isinstance(target, ManagerTarget) or (
+            isinstance(target, str) and target in {"manager", "secondary_manager", "tertiary_manager"}
+        ):
+            # The bare-string manager targets and the ``{manager: true}``
+            # dict form (fixture 036, proposal 0086) both dispatch to the
+            # currently-active manager in the per-pair iteration loop. The
+            # string naming keeps fixture YAML self-describing under a
             # multi-manager shape (e.g., fixture 015).
             assert manager is not None
             if operation == "fetch":
@@ -392,7 +397,7 @@ async def _run_call(
             raise AssertionError(f"unsupported manager operation: {operation!r}")
 
         # ``target: {backend: <name>}`` — direct backend op.
-        assert not isinstance(target, str)
+        assert isinstance(target, BackendTarget)
         backend = backends[target.backend]
         if operation == "fetch":
             assert call.name is not None and call.label is not None
@@ -528,13 +533,23 @@ def _build_manager(
     backends_map: dict[str, MockPromptBackend],
     resolvers_map: dict[str, MappingLabelResolver],
 ) -> PromptManager:
-    ordered = [backends_map[name] for name in spec.backends]
+    # A manager spec that omits ``backends`` uses all declared backends in
+    # declaration order (fixture 036's single-backend default-cache-ttl shape).
+    ordered = (
+        [backends_map[name] for name in spec.backends]
+        if spec.backends is not None
+        else list(backends_map.values())
+    )
     resolver: MappingLabelResolver | None = None
     if spec.label_resolver_ref is not None:
         if spec.label_resolver_ref not in resolvers_map:
             raise AssertionError(f"unknown label_resolver_ref: {spec.label_resolver_ref!r}")
         resolver = resolvers_map[spec.label_resolver_ref]
-    return PromptManager(*ordered, label_resolver=resolver)
+    return PromptManager(
+        *ordered,
+        label_resolver=resolver,
+        default_cache_ttl_seconds=spec.default_cache_ttl_seconds,
+    )
 
 
 def _assert_capture_attrs(capture_name: str, actual: Any, expected: dict[str, Any]) -> None:
@@ -611,12 +626,6 @@ def _message_to_dict_for_compare(message: Message) -> dict[str, Any]:
     # accept, and asserts the construct-time prompt_group_invalid raise that
     # python does not yet implement. Defers until a later v0.16.0 PR.
     "035-prompt-group-arity-rejection": ("Proposal 0080 PromptGroup arity enforcement; not implemented"),
-    # Proposal 0086 (PromptManager default cache_ttl_seconds, spec v0.79.0)
-    # -- fixture 036 uses the manager default-cache-ttl construction slot
-    # python does not yet implement. Defers until a later v0.16.0 PR.
-    "036-prompt-manager-default-cache-ttl": (
-        "Proposal 0086 PromptManager default cache_ttl_seconds; not implemented"
-    ),
 }
 
 

tests/unit/test_prompts.py

@@ -268,6 +268,87 @@ async def fetch(
         await manager.fetch("greeting", "production", cache_ttl_seconds=-1)
 
 
+# ---------------------------------------------------------------------------
+# Proposal 0086: PromptManager service-wide default_cache_ttl_seconds
+# ---------------------------------------------------------------------------
+
+
+class _RecordingCacheTtlBackend:
+    # Records the cache_ttl_seconds the manager forwarded on the last fetch,
+    # so the resolution precedence can be asserted at the backend boundary.
+    def __init__(self, prompt: Prompt) -> None:
+        self._prompt = prompt
+        self.last_cache_ttl_seconds: int | None = None
+
+    async def fetch(
+        self, name: str, label: str = "production", *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
+        self.last_cache_ttl_seconds = cache_ttl_seconds
+        return self._prompt
+
+
+def test_construction_rejects_negative_default_cache_ttl() -> None:
+    # A negative service-wide default is invalid and is rejected at
+    # construction, before any fetch.
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    with pytest.raises(ValueError, match="default_cache_ttl_seconds must be >= 0"):
+        PromptManager(backend, default_cache_ttl_seconds=-1)
+
+
+async def test_default_cache_ttl_applies_when_per_call_omitted() -> None:
+    # Precedence step 2: an omitted per-call value resolves to the manager
+    # default, which is forwarded to the backend verbatim.
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    manager = PromptManager(backend, default_cache_ttl_seconds=60)
+    await manager.fetch("greeting", "production")
+    assert backend.last_cache_ttl_seconds == 60
+
+
+async def test_explicit_none_selects_default_cache_ttl() -> None:
+    # Resolution is presence-independent: an explicit None selects the
+    # default exactly as an omitted argument does.
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    manager = PromptManager(backend, default_cache_ttl_seconds=60)
+    await manager.fetch("greeting", "production", cache_ttl_seconds=None)
+    assert backend.last_cache_ttl_seconds == 60
+
+
+async def test_per_call_zero_overrides_positive_default_cache_ttl() -> None:
+    # Precedence step 1: an explicit per-call value wins, so a 0 force-fresh
+    # overrides a positive manager default.
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    manager = PromptManager(backend, default_cache_ttl_seconds=60)
+    await manager.fetch("greeting", "production", cache_ttl_seconds=0)
+    assert backend.last_cache_ttl_seconds == 0
+
+
+async def test_no_default_forwards_none_cache_ttl() -> None:
+    # Precedence step 3: with no manager default and no per-call value, None
+    # is forwarded, so the backend's own caching governs (unchanged).
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    manager = PromptManager(backend)
+    await manager.fetch("greeting", "production")
+    assert backend.last_cache_ttl_seconds is None
+
+
+async def test_zero_default_cache_ttl_is_valid_and_forwarded() -> None:
+    # A default of 0 is valid (force-fresh-always): accepted at construction,
+    # and an omitted per-call value resolves to 0, forwarded to the backend.
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    manager = PromptManager(backend, default_cache_ttl_seconds=0)
+    await manager.fetch("greeting", "production")
+    assert backend.last_cache_ttl_seconds == 0
+
+
+async def test_get_inherits_default_cache_ttl() -> None:
+    # get() delegates to fetch(), so the manager default resolves for get()
+    # too when the per-call value is omitted.
+    backend = _RecordingCacheTtlBackend(_make_prompt())
+    manager = PromptManager(backend, default_cache_ttl_seconds=60)
+    await manager.get("greeting", "production", {"user": "Alice"})
+    assert backend.last_cache_ttl_seconds == 60
+
+
 # ---------------------------------------------------------------------------
 # FilesystemPromptBackend
 # ---------------------------------------------------------------------------