Add per-fetch prompt cache control cache_ttl_seconds (0072)

CHANGELOG.md

@@ -11,10 +11,11 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). The
 - **Detached-trace invocation span** (proposal 0061, observability §4.4, spec v0.61.0). The OTel observer now synthesizes an `openarmature.invocation` span at the root of each detached trace (a detached subgraph and each detached fan-out instance), carrying the parent's shared `invocation_id` (detached mode is observer-side trace rendering, not a new run) and the detached unit's own `entry_node`; the detached subgraph / instance span nests under it. A raising detached subgraph surfaces ERROR plus the error category and an OTel exception event on both the parent dispatch span and the detached invocation span. This is observer-side only, with no graph-engine change; the Langfuse observer is unchanged (its Trace entity already plays the invocation-level-container role). Conformance fixtures 008 (rewritten) and 058 (newly wired) run in `test_observability`.
 - **Per-attempt LLM spans under call-level retry** (proposal 0050, observability §5.5 / llm-provider §7.1). Completes proposal 0050, which shipped `partial` in v0.14.0 (failure-isolation middleware and the `complete(retry=...)` loop landed then; the per-attempt span surface was deferred). Under call-level retry the OTel observer now emits one `openarmature.llm.complete` span per attempt, each carrying `openarmature.llm.attempt_index` (0-based, 0..N-1, and 0 for a no-retry call). An intermediate failed attempt's span carries ERROR status plus its error category and the request-side attributes; the final attempt's span carries the terminal outcome and, on success, the full response surface. A python-internal `LlmRetryAttemptEvent`, dispatched once per attempt, is the sole source of the OTel span; the terminal `LlmCompletionEvent` / `LlmFailedEvent` stay one per call (payload, latency, Langfuse Generation) and no longer drive the OTel span. Langfuse renders one terminal Generation per call, with the per-attempt detail on the OTel span surface only (a spec-side §8 clarification to pin this is tracked, non-blocking). `conformance.toml` flips proposal 0050 to `implemented`; the call-level fixtures 056-058 are driven through the provider plus OTel observer and the single-attempt observability fixture 057 is wired.
 - **Langfuse `trace.userId` / `trace.sessionId` population** (proposal 0064, observability §8.4.1, spec v0.62.0). The Langfuse observer now promotes a recognized `userId` key in the caller-supplied invocation metadata to Langfuse's first-class `trace.userId` field (the Users dashboard), additively: the key also remains at `trace.metadata.userId`. Promotion is automatic and unconditional; an absent key leaves `trace.userId` unset. The `LangfuseClient.trace()` surface (the Protocol, the in-memory client, and the SDK adapter) gains `session_id` / `user_id`. `trace.sessionId` is sourced from `openarmature.session_id`, which the sessions capability (proposal 0020) establishes; that capability is not yet implemented in python, so the `sessionId` plumbing is in place but dormant (no source) and unset in the interim. `conformance.toml` records proposal 0064 `partial` on that basis: fixture 084 cases 2/3/4 (not session-bound, `userId` present additively, `userId` absent) run, and the session-bound cases 1/5 defer until 0020. Langfuse-only: the OTel side already carries `openarmature.session_id` and `openarmature.user.*` as span attributes, and OTel has no trace-level session/user field.
+- **Per-fetch prompt cache control: `cache_ttl_seconds`** (proposal 0072, prompt-management §5 / §6, spec v0.63.0). `PromptBackend.fetch`, `PromptManager.fetch`, and `PromptManager.get` gain an optional `cache_ttl_seconds` read-side control: `None` preserves current behavior, `0` forces a fresh read past any client-side cache, and `N > 0` bounds a served entry's staleness to N seconds; a negative value is rejected at the manager. It governs only which cached entry may be served, not whether or how results are cached. The bundled filesystem backend is cacheless and ignores it; the bundled Langfuse backend forwards it to the Langfuse SDK's `get_prompt` cache. Conformance fixtures 033/034 run through a caching harness backend (conformance-adapter §6.8: `source_read_count` plus a controllable `advance_clock`).
 
 ### Changed
 
-- **Pinned spec advances v0.60.0 → v0.62.0** across the v0.15.0 cycle: v0.61.0 (proposal 0061, the detached-trace invocation span above) and v0.62.0 (proposal 0064, the Langfuse session/user population above). `conformance.toml` records 0061 `implemented` and 0064 `partial` (its `sessionId` half is dormant pending the sessions capability). Proposal 0050 needed no pin bump of its own (it was already within the pin from its v0.42.0 acceptance); its v0.14.0 `partial` entry flips to `implemented` with the per-attempt span surface above.
+- **Pinned spec advances v0.60.0 → v0.63.1** across the v0.15.0 cycle: v0.61.0 (proposal 0061, the detached-trace invocation span above), v0.62.0 (proposal 0064, the Langfuse session/user population above), v0.63.0 (proposal 0072, the prompt cache control above), and the v0.63.1 patch (pipeline-utilities coverage fixtures 070/071 for the already-implemented 0069 / 0070 behavior, no new proposal). `conformance.toml` records 0061 / 0072 `implemented` and 0064 `partial` (its `sessionId` half is dormant pending the sessions capability). Proposal 0050 needed no pin bump of its own (it was already within the pin from its v0.42.0 acceptance); its v0.14.0 `partial` entry flips to `implemented` with the per-attempt span surface above.
 
 ## [0.14.0] — 2026-06-17
 

conformance.toml

@@ -705,3 +705,10 @@ note = "The OTel observer synthesizes an openarmature.invocation span at the roo
 status = "partial"
 since = "0.15.0"
 note = "The Langfuse observer promotes a recognized userId caller-metadata key to the first-class trace.userId (additive: the key also stays in trace.metadata.userId), and sets trace.sessionId from openarmature.session_id when present. trace.userId is LIVE (sourced from 0034 caller metadata): fixture 084 cases 2/3/4 (not-session-bound, userId present additive, userId absent) pass. partial because trace.sessionId is DORMANT -- openarmature.session_id is established by the sessions capability (0020, observability §5.6), unimplemented in python until v0.19.0, so there is no session_id source yet; the trace(session_id=) plumbing is wired end to end but the observer passes None. Fixture 084 session-bound cases 1 + 5 are deferred (per-case) pending 0020. Langfuse-only: no OTel change (the OTel side already carries openarmature.session_id + openarmature.user.* as span attributes; no trace-level OTel equivalent)."
+
+# Spec v0.63.0 (proposal 0072).  Per-fetch cache_ttl_seconds read-side
+# control (prompt-management §5 / §6 + conformance-adapter §6.8).
+[proposals."0072"]
+status = "implemented"
+since = "0.15.0"
+note = "PromptBackend.fetch / PromptManager.fetch / get gain an optional cache_ttl_seconds read-side control (absent / None = current behavior; 0 = force a fresh read past any cache; N > 0 = bound a served entry's staleness to N seconds; negative is rejected). It governs only which cached entry MAY be served for this fetch, not whether / how the result is cached. python's bundled backends (filesystem, in-memory) are cacheless and treat it as a no-op; the manager threads it through the §9 fallback chain and rejects negatives. render is unchanged. The TTL semantics are exercised by a caching prompt-backend conformance-harness primitive (§6.8: caches by (name, label), source_read_count, advance_clock controllable clock); fixtures 033/034 pass. No production caching backend ships (per §5, cacheless backends no-op). The v0.63.1 pin also wires pipeline-utilities coverage fixtures 070/071 (already-implemented 0069/0070 behavior; no new proposal)."

docs/concepts/prompts.md

@@ -47,6 +47,32 @@ Why two operations instead of one? Three reasons:
 The convenience `get()` operation gives you the single-call
 shape when you want it without removing the separability.
 
+## Refreshing cached prompts: `cache_ttl_seconds`
+
+`fetch` and `get` take an optional `cache_ttl_seconds` that controls how
+fresh a served prompt must be, for backends that maintain a client-side
+cache:
+
+- omitted / `None` keeps the backend's current behavior;
+- `0` forces a fresh read past any cache;
+- `N > 0` serves a cached entry only while it is younger than N seconds,
+  re-reading the source once it ages past N.
+
+A negative value is rejected. It is a read-side control: it governs which
+cached entry may be served for this fetch, not whether or how results are
+cached. Cacheless backends (the bundled filesystem backend) ignore it; the
+bundled Langfuse backend forwards it to the Langfuse SDK's own prompt cache.
+
+```python
+# Always re-read from the backend, bypassing any cache:
+fresh = await manager.fetch("greeting", "production", cache_ttl_seconds=0)
+
+# Serve a cached entry only if it's under five minutes old:
+recent = await manager.get(
+    "greeting", "production", {"user": "Alice"}, cache_ttl_seconds=300
+)
+```
+
 ## Prompt identity
 
 Every `Prompt` carries five identity fields:

examples/chat-with-multimodal/main.py

@@ -274,7 +274,9 @@ class _NoFetchBackend:
     ``fetch()`` is never invoked.
     """
 
-    async def fetch(self, name: str, label: str = "production") -> Prompt:
+    async def fetch(
+        self, name: str, label: str = "production", *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
         raise NotImplementedError("example constructs prompts inline; fetch not used")
 
 

examples/langfuse-observability/main.py

@@ -123,7 +123,9 @@ def __init__(self) -> None:
             },
         )
 
-    async def fetch(self, name: str, label: str = "production") -> Prompt:
+    async def fetch(
+        self, name: str, label: str = "production", *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
         if name != "mission-briefing":
             from openarmature.prompts import PromptNotFound
 

pyproject.toml

@@ -63,7 +63,7 @@ Specification = "https://github.com/LunarCommand/openarmature-spec"
 openarmature = "openarmature.cli:main"
 
 [tool.openarmature]
-spec_version = "0.62.0"
+spec_version = "0.63.1"
 
 [dependency-groups]
 dev = [

src/openarmature/AGENTS.md

@@ -1,6 +1,6 @@
 # OpenArmature — Agent documentation
 
-*This is the agent guide bundled with the openarmature Python package, version 0.14.0 (spec v0.62.0). For the full docs site see [openarmature.ai](https://openarmature.ai). For the canonical spec text see [openarmature.org/capabilities](https://openarmature.org/capabilities/). For project-specific conventions for the code you're editing, see the host project's `AGENTS.md` or `CLAUDE.md`.*
+*This is the agent guide bundled with the openarmature Python package, version 0.14.0 (spec v0.63.1). For the full docs site see [openarmature.ai](https://openarmature.ai). For the canonical spec text see [openarmature.org/capabilities](https://openarmature.org/capabilities/). For project-specific conventions for the code you're editing, see the host project's `AGENTS.md` or `CLAUDE.md`.*
 
 ## TL;DR
 
@@ -10,7 +10,7 @@ OpenArmature is a workflow framework for LLM pipelines and tool-calling agents:
 
 ## Capability contracts
 
-_Sourced from openarmature-spec v0.62.0. Each entry below reproduces §1 (Purpose) and §2 (Concepts) of the capability's `spec.md` verbatim — including additions from accepted proposals that this Python implementation may not yet ship. For per-proposal implementation status (implemented / partial / textual-only / not-yet), see the `conformance.toml` manifest at the repo root. For the full spec text (execution model, error semantics, determinism, observer hooks, etc.) see the linked docs site._
+_Sourced from openarmature-spec v0.63.1. Each entry below reproduces §1 (Purpose) and §2 (Concepts) of the capability's `spec.md` verbatim — including additions from accepted proposals that this Python implementation may not yet ship. For per-proposal implementation status (implemented / partial / textual-only / not-yet), see the `conformance.toml` manifest at the repo root. For the full spec text (execution model, error semantics, determinism, observer hooks, etc.) see the linked docs site._
 
 ### Capability: `graph-engine`
 

src/openarmature/__init__.py

@@ -25,7 +25,7 @@
 """
 
 __version__ = "0.14.0"
-__spec_version__ = "0.62.0"
+__spec_version__ = "0.63.1"
 # Proposal 0052 (spec observability §5.1 / §8.4.1): canonical
 # package-registry name for this implementation. Surfaces on every
 # OTel invocation span as ``openarmature.implementation.name`` and on

src/openarmature/prompts/backend.py

@@ -33,13 +33,21 @@ class PromptBackend(Protocol):
     original fetch time, not the cache hit time.
     """
 
-    async def fetch(self, name: str, label: str = "production") -> Prompt:
+    async def fetch(
+        self, name: str, label: str = "production", *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
         """Return the prompt registered as ``(name, label)``.
 
         ``label`` defaults to ``"production"``. Raises
         ``PromptNotFound`` if no prompt matches, and
         ``PromptStoreUnavailable`` if the backing store is unreachable.
         The returned ``Prompt`` carries its raw template plus
         metadata; rendering is the manager's job, not the backend's.
+
+        ``cache_ttl_seconds`` is a read-side cache control: ``None``
+        preserves the backend's current behavior, ``0`` forces a fresh
+        read past any client-side cache, and ``N > 0`` bounds a served
+        cached entry's staleness to N seconds. Cacheless backends ignore
+        it; caching backends honor it.
         """
         ...

src/openarmature/prompts/backends/filesystem.py

@@ -143,13 +143,18 @@ def _resolve_sampling(self, name: str, label: str) -> SamplingConfig | None:
             )
         return _sampling_from_dict(cast(dict[str, Any], raw))
 
-    async def fetch(self, name: str, label: str = "production") -> Prompt:
+    async def fetch(
+        self, name: str, label: str = "production", *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
         """Read the prompt template and (optionally) its sidecar sampling config.
 
         Returns a ``Prompt`` whose ``version`` is the leading 16 hex
         chars of the template's SHA-256 and ``template_hash`` is the
         full digest. Raises ``PromptNotFound`` when the template is
         missing and ``PromptStoreUnavailable`` on other I/O errors.
+
+        The filesystem backend is cacheless, so ``cache_ttl_seconds`` is
+        accepted for protocol conformance and ignored.
         """
         path = self._template_path(name, label)
         try:

src/openarmature/prompts/backends/langfuse.py

@@ -46,7 +46,9 @@ class LangfusePromptClient(Protocol):
     tests can supply a lightweight fake.
     """
 
-    def get_prompt(self, name: str, *, label: str = "production") -> TextPromptClient | ChatPromptClient: ...
+    def get_prompt(
+        self, name: str, *, label: str = "production", cache_ttl_seconds: int | None = None
+    ) -> TextPromptClient | ChatPromptClient: ...
 
 
 # Langfuse prompt `config` keys that line up with SamplingConfig's
@@ -89,10 +91,14 @@ class LangfusePromptBackend:
     def __init__(self, client: LangfusePromptClient) -> None:
         self._client = client
 
-    async def fetch(self, name: str, label: str = "production") -> Prompt:
+    async def fetch(
+        self, name: str, label: str = "production", *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
         # The Langfuse SDK's get_prompt is synchronous (and does its own
-        # client-side caching); run it off the event loop.
-        result = await asyncio.to_thread(self._get_prompt, name, label)
+        # client-side caching); run it off the event loop. The proposal
+        # 0072 cache_ttl_seconds control forwards to that SDK cache:
+        # None = SDK default, 0 = no cache (fresh), N = N-second bound.
+        result = await asyncio.to_thread(self._get_prompt, name, label, cache_ttl_seconds)
 
         if isinstance(result, ChatPromptClient):
             normalized = _normalized_langfuse_entries(result.prompt, name=name, label=label)
@@ -134,9 +140,11 @@ async def fetch(self, name: str, label: str = "production") -> Prompt:
             metadata=_metadata_from(result),
         )
 
-    def _get_prompt(self, name: str, label: str) -> TextPromptClient | ChatPromptClient:
+    def _get_prompt(
+        self, name: str, label: str, cache_ttl_seconds: int | None = None
+    ) -> TextPromptClient | ChatPromptClient:
         try:
-            return self._client.get_prompt(name, label=label)
+            return self._client.get_prompt(name, label=label, cache_ttl_seconds=cache_ttl_seconds)
         except NotFoundError as exc:
             raise PromptNotFound(
                 f"prompt ({name!r}, {label!r}) not found in Langfuse",

src/openarmature/prompts/manager.py

@@ -110,7 +110,9 @@ def _resolve_label(self, label: str | None, name: str) -> str:
             return self._label_resolver.resolve(name)
         return SPEC_FALLBACK_LABEL
 
-    async def fetch(self, name: str, label: str | None = None) -> Prompt:
+    async def fetch(
+        self, name: str, label: str | None = None, *, cache_ttl_seconds: int | None = None
+    ) -> Prompt:
         """Consult composed backends in order, applying the fallback chain.
 
         Label is resolved by a three-step chain: explicit argument >
@@ -123,12 +125,23 @@ async def fetch(self, name: str, label: str | None = None) -> Prompt:
         - ``PromptStoreUnavailable`` from a backend continues to the
           next. After ALL backends are exhausted with unavailable
           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.
         """
+        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"
+            )
         resolved_label = self._resolve_label(label, name)
         causes: list[BaseException] = []
         for backend in self._backends:
             try:
-                return await backend.fetch(name, resolved_label)
+                return await backend.fetch(name, resolved_label, cache_ttl_seconds=cache_ttl_seconds)
             except PromptNotFound:
                 raise
             except PromptStoreUnavailable as exc:
@@ -520,13 +533,16 @@ async def get(
         variables: Mapping[str, Any] | None = None,
         *,
         placeholders: Mapping[str, Sequence[Message]] | None = None,
+        cache_ttl_seconds: int | None = None,
     ) -> PromptResult:
         """Convenience equivalent to ``render(await fetch(name, label), variables)``.
 
         ``label`` follows the same three-step resolution as :meth:`fetch`.
         ``placeholders`` is forwarded to :meth:`render`.
+        ``cache_ttl_seconds`` is forwarded to :meth:`fetch` (the read-side
+        cache control).
         """
-        prompt = await self.fetch(name, label)
+        prompt = await self.fetch(name, label, cache_ttl_seconds=cache_ttl_seconds)
         return self.render(prompt, variables, placeholders=placeholders)