Close durable runtime storage and tracing audit

coding-agent/.opencode/prompts/tasks/durable-runtime-g03-g08-audit.md

@@ -0,0 +1,57 @@
+Goal:
+Audit and close active durable runtime G03 and G08 progress gaps.
+
+Scope:
+- Confirm G03 storage/plugin/composition wiring preserves JSONL defaults and
+  only enables PostgreSQL runtime storage when `storage.runtime_backend = "pg"`.
+- Confirm G08 observability correlation propagates allowed runtime identity
+  keys and rejects unsafe raw prompt/content/message/result/secret/text-style
+  attributes.
+- Add focused test coverage only where the audit evidence is too implicit.
+- Update `docs/durable_runtime/GOAL_PROGRESS.md`.
+
+Out of scope:
+- Do not change production storage behavior.
+- Do not change OTLP/Langfuse exporter behavior.
+- Do not introduce new runtime stores, endpoints, or migrations.
+- Do not work on active G09 tape debug APIs or active G11 smoke docs/tests in
+  this slice.
+
+Context:
+- ADRs:
+  - `docs/adr/0029-durable-runtime-identity.md`
+  - `docs/adr/0030-postgresql-durable-runtime-store.md`
+  - `docs/adr/0032-durable-runtime-lifecycle-statuses.md`
+- Postmortem patterns consulted:
+  - `postmortem/patterns/PM-0001-address-code-review-issues.md`
+  - `postmortem/patterns/PM-0006-add-usage-event-fields-and-fix-tool-name-kwarg-in-pipeline.md`
+  - `postmortem/patterns/PM-0009-preserve-neutral-bare-anchor-semantics.md`
+  - `postmortem/patterns/PM-0010-route-incremental-context-append-through-tapeview.md`
+- Relevant files:
+  - `src/coding_agent/ui/session_manager.py`
+  - `src/coding_agent/observability.py`
+  - `src/agentkit/runtime/pipeline.py`
+  - `tests/ui/test_session_manager_public_api.py`
+  - `tests/agentkit/runtime/test_pipeline.py`
+  - `tests/coding_agent/test_observability.py`
+  - `docs/durable_runtime/GOAL_PROGRESS.md`
+
+Target tests:
+- `uv run pytest tests/ui/test_session_manager_public_api.py -k "storage_config or runtime_store or pg_backends or pg_pool or runtime_backend" -v`
+- `uv run pytest tests/agentkit/runtime/test_pipeline.py -k "span" -v`
+- `uv run pytest tests/coding_agent/test_observability.py -v`
+- `uv run ruff check tests/agentkit/runtime/test_pipeline.py`
+- `uv run ruff format --check tests/agentkit/runtime/test_pipeline.py`
+
+Loop policy:
+- Engineer implements the smallest evidence/coverage change and runs target
+  tests.
+- Reviewer reviews only the resulting diff and affected tests.
+- Reviewer reports only P1/P2 findings.
+- Engineer fixes only accepted P1/P2 findings and reruns the same target tests.
+- Verifier reruns the exact target tests and reports pass/fail only.
+
+Stop conditions:
+- Stop with a blocker if G03 requires production storage changes.
+- Stop with a blocker if G08 requires weakening the observability denylist.
+- Stop with a blocker if this audit expands into G09 tape debug work.

coding-agent/docs/durable_runtime/GOAL_PROGRESS.md

@@ -13,12 +13,12 @@ match the active objective map.
 | G00 | Complete | `docs/durable_runtime/CURRENT_STATE.md` documents SessionManager, PipelineAdapter, storage plugins, PG storage, observability, approval flow, event flow, and tape flow. |
 | G01 | Complete | `docs/adr/0029-durable-runtime-identity.md` defines session, run, turn compatibility, tape, event, interaction, checkpoint, and Langfuse/OTLP correlation identity. |
 | G02 | Complete | `src/coding_agent/runtime_store.py` adds the PostgreSQL durable runtime store, with unit coverage in `tests/coding_agent/test_pg_runtime_store.py` and ADR-0030. |
-| G03 | Needs audit | `storage.runtime_backend = "pg"` opt-in wiring exists, but the storage/plugin/composition boundary still needs final audit against the active objective wording. |
+| G03 | Complete | Storage configuration preserves JSONL/no-runtime defaults and opt-in PG composition wires tape, checkpoint, HTTP session, and runtime stores through tests in `tests/ui/test_session_manager_public_api.py`. |
 | G04 | Complete | ADR-0032 and `SessionManager` now persist `queued -> running -> completed/failed/cancelled/interrupted` root run lifecycle rows while preserving `current_turn_id` compatibility. |
 | G05 | Complete | `SessionManager` persists runtime wire events and `{run_id}:latest` compacted message snapshots when a runtime store is configured. |
 | G06 | Complete | HTTP replay APIs expose runtime run records, latest message snapshots, and runtime events with `last_event_id` filtering. |
 | G07 | Complete | `SessionManager` persists durable approval interaction records for pending requests, applied decisions, session auto-approvals, and approval timeouts when a runtime store is configured. |
-| G08 | Needs audit | AgentKit and HTTP root run correlation were added, but final audit must confirm the active objective's full safe attribute set is populated where required. |
+| G08 | Complete | AgentKit stage/LLM spans include safe runtime correlation keys (`turn_id`, `tape_id`, `tool_call_id`, `interaction_id`, `event_id`, `checkpoint_id`), HTTP root runs bind `turn_id`/`tape_id`, and OTLP privacy tests reject raw prompt/message/result/secret/text attributes. |
 | G09 | Pending | `tape.info` and `tape.search` for PostgreSQL tape storage are still missing. |
 | G10 | Complete | ADR-0032 changes startup orphan recovery to mark active-owner `running` rows `interrupted` with `reclaimable: true`, without adding a scheduler. |
 | G11 | Pending | End-to-end smoke tests/docs for normal run, failed run, approval run, replay, Langfuse/OTLP correlation, and tape debug remain to be added. |
@@ -48,6 +48,24 @@ match the active objective map.
   - `postmortem/patterns/PM-0022-revalidate-event-stream-ownership-after-queue-attach.md`
   - `postmortem/patterns/PM-0023-make-event-stream-cleanup-and-teardown-idempotent.md`
 
+## Active G03/G08 Audit Verification
+
+- Added `.opencode/prompts/tasks/durable-runtime-g03-g08-audit.md`.
+- Updated `tests/agentkit/runtime/test_pipeline.py` to assert all allowed
+  runtime correlation metadata keys are propagated while unsafe/non-scalar
+  metadata is dropped.
+- Updated this progress table to move active G03 and G08 from `Needs audit` to
+  `Complete`.
+- Baseline and target tests:
+  - `uv run pytest tests/ui/test_session_manager_public_api.py -k "storage_config or runtime_store or pg_backends or pg_pool or runtime_backend" -v`
+  - `uv run pytest tests/agentkit/runtime/test_pipeline.py -k "span" -v`
+  - `uv run pytest tests/coding_agent/test_observability.py -v`
+- Postmortem patterns consulted:
+  - `postmortem/patterns/PM-0001-address-code-review-issues.md`
+  - `postmortem/patterns/PM-0006-add-usage-event-fields-and-fix-tool-name-kwarg-in-pipeline.md`
+  - `postmortem/patterns/PM-0009-preserve-neutral-bare-anchor-semantics.md`
+  - `postmortem/patterns/PM-0010-route-incremental-context-append-through-tapeview.md`
+
 ## Historical Slice Log
 
 ## G00 Verification

coding-agent/tests/agentkit/runtime/test_pipeline.py

@@ -386,8 +386,16 @@ async def mock_stream(messages, tools=None, **kwargs):
                 context_budget=ContextBudget(),
                 trace_metadata={
                     "turn_id": "run-1",
+                    "tool_call_id": "tool-call-1",
                     "interaction_id": "interaction-1",
+                    "event_id": "event-1",
+                    "checkpoint_id": "checkpoint-1",
                     "prompt": "do-not-record",
+                    "message": "do-not-record",
+                    "result": "do-not-record",
+                    "secret": "do-not-record",
+                    "text": "do-not-record",
+                    "ignored_object": {"nested": "do-not-record"},
                 },
             ),
             config={"observation_sink": sink},
@@ -416,11 +424,21 @@ async def mock_stream(messages, tools=None, **kwargs):
             "agent_id": "root",
             "tape_id": "tape-1",
             "turn_id": "run-1",
+            "tool_call_id": "tool-call-1",
             "interaction_id": "interaction-1",
+            "event_id": "event-1",
+            "checkpoint_id": "checkpoint-1",
             "entry_count_before": 0,
             "entry_count_after": 0,
         }
-        assert "prompt" not in build_context_span.attributes
+        assert {
+            "prompt",
+            "message",
+            "result",
+            "secret",
+            "text",
+            "ignored_object",
+        }.isdisjoint(build_context_span.attributes)
         model_span = stage_spans[3]
         assert model_span.attributes["entry_count_before"] == 0
         assert model_span.attributes["entry_count_after"] == 1