Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cmd/seed-builtin-skills/main.go`:
- Around line 70-74: Update the seed-result reporting after the import operation
to use the injected structured logger instead of the three fmt.Printf calls. Log
the imported count, optional pruned count, and result.Skills as separate
structured fields while preserving the existing message content and conditional
pruned reporting.

In `@internal/agents/handler.go`:
- Line 174: Preserve context and resource scope in every listed structured
failure log: in internal/agents/handler.go lines 174, 219, 261, 302, 341, 380,
406, and 444, replace logger Error calls with ErrorContext(r.Context(), ...) and
add the specified agent_id or workspace_id; update lines 1237-1243 to use
ErrorContext(ctx, ...). In internal/skillprewarm/worker.go lines 98 and 194, use
ErrorContext(ctx, ...).

In `@internal/api/http_log_test.go`:
- Around line 58-60: Strengthen the assertion in the HTTP logging test around
the buffer check so it verifies that no query values are logged, including both
“secret-state” and “beta=true”. Prefer rejecting the request’s complete RawQuery
representation when validating the no-query logging contract.

In `@internal/api/platform_mcp_vault_auth.go`:
- Around line 201-203: Sanitize all OAuth URL/query-derived values before
logging in internal/api/platform_mcp_vault_auth.go: at lines 201-203, replace
the raw discovery error with allow-listed metadata such as host and HTTP status;
at lines 337-339, omit the callback’s raw error query value and log only a fixed
event and flow ID; at lines 482-488, similarly sanitize protected-resource
metadata fetch errors before attaching them to logs. Ensure no complete query
strings, OAuth codes/state, credentials, or credential payloads are recorded.

In `@internal/api/server.go`:
- Around line 277-283: Update the panic-recovery logging around the ErrorContext
call to stop recording the arbitrary recovered panic value, which may contain
sensitive request-derived data. Preserve logging of request_id and stack, and
remove the "panic" field or replace it only with an approved sanitized
representation.

In `@internal/codesessions/ingress.go`:
- Around line 756-763: Update internal/codesessions/ingress.go lines 756-763:
extend Handler.writeIngressLoadError with the available codeSessionID parameter,
update all callers to pass it, and include it as "code_session_id" in the
internal-error log. Update internal/sessions/event_payload.go lines 418-429,
431-438, and 440-447: add "session_id", "thread_id", and "resource_id"
respectively to the internal-error logs in writeSessionLoadError,
writeThreadLoadError, and writeResourceLoadError, reusing their existing
identifier parameters.
- Around line 55-57: Replace contextless structured logging with the
context-aware slog methods wherever context is available. In
internal/codesessions/ingress.go at 55-57, 65-67, 77-79, 110-112, 172-180,
201-203, 251-253, 305-313, 328-330, 367-369, 398-400, 433-435, 460-462, 474-476,
535-537, 563-565, 595-597, 760-762, and 836-838, use ErrorContext or WarnContext
with r.Context(), or the existing ctx parameter in
streamCodeSessionWorkerEvents; in internal/sessions/event_effects.go at 23-31,
46-52, 63-70, 86, 99, and 113, use ErrorContext(ctx, ...); in
internal/sessions/event_payload.go at 400-416, 418-429, 431-438, and 440-447,
use ErrorContext(r.Context(), ...); and in
internal/codesessions/otlp_file_log.go at 126-128, use ErrorContext(r.Context(),
...), preserving each existing log message and attributes.

In `@internal/codesessions/service.go`:
- Around line 37-43: Run go test ./... -count=1 and fix the failures in
internal/workbench and the Postgres-backed test suite. Update the affected test
setup and callers of NewServiceWithCredentials to provide valid
SessionCredentials and required Postgres configuration, preserving the
constructor’s non-nil credential contract. Verify the full test suite passes.

In `@internal/filestore/handler.go`:
- Around line 94-100: The filestore panic recovery log must not include the raw
recovered value. In the panic handler containing the “filestore panic recovered”
message, remove the “panic” attribute while preserving the request ID and stack
fields, and add a structured-record test that verifies a sensitive panic value
is absent from the emitted log.

In `@internal/sessions/transport.go`:
- Around line 18-21: Change NewHandler to return (*Handler, error) and replace
the nil codeSessionService panic with a descriptive construction error. Update
every caller, especially the executable startup boundary, to propagate or handle
the returned error so the composition layer can log the terminal startup failure
and choose the exit code.

In `@internal/workbench/workbench_handler.go`:
- Line 62: Update the POST revision rename route to call the dedicated revision
rename handler instead of handleGetWorkbenchPromptRevision. Preserve the
existing route path and method, and use the handler symbol already defined for
renaming revisions.

In `@main.go`:
- Line 95: Guard the batches.NewWorker(...).Start(ctx) call with
cfg.Batch.WorkerEnabled so batch processing starts only when the worker is
enabled. Preserve the existing worker construction and logger configuration when
the gate is true.

In `@tests/files_api_test.go`:
- Around line 452-454: Move the capture logger setup in the test before
constructing fakeApp, preserving the original logger for deferred restoration.
Ensure api.NewServer receives and scopes the capture logger so the logs buffer
records the expected stream-error entry.

---

Outside diff comments:
In `@internal/api/http_log_test.go`:
- Around line 42-50: Rename the HTTP log attributes emitted by the request
logging implementation and asserted in the test from camelCase to snake_case:
requestId to request_id, clientKind to client_kind, anthropicClientPlatform to
anthropic_client_platform, anthropicClientApp to anthropic_client_app, and
durationMs to duration_ms. Update the relevant logging code in http_log.go and
matching assertions in http_log_test.go consistently.

In `@internal/api/http_log.go`:
- Around line 24-40: Update the structured attribute names produced by
httpLogAttrs and the response attributes in the HTTP logging middleware to use
snake_case, including request_id, duration_ms, user_agent, client_kind, and all
Anthropic-related fields. Preserve the existing attribute values and logging
behavior.

In `@internal/api/server.go`:
- Around line 84-142: Run just dead-code, just lint, just duplicates, and go
test ./... -count=1 for the NewServer routing and handler changes, then address
any reported issues before merging. Ensure the PR summary records the checks and
does not state that tests were skipped.

In `@internal/memory/handler.go`:
- Around line 1745-1808: Update writeStoreLoadError, writeMemoryLoadError,
writeVersionLoadError, and writeMemoryMutationError to include their available
resource IDs as structured attributes (memory_store_id, memory_id, and
memory_version_id) in failure logs while preserving the request context; add a
structured logging test that verifies the expected message and fields.

In `@internal/workbench/console_platform_workbench.go`:
- Around line 1952-1979: Make the multi-record persistence in storeRevision and
the corresponding storeKV flow atomic by introducing or reusing a store
operation that upserts the related WorkbenchRevisionRecord/WorkbenchPromptRecord
or prompt/KV records within one database transaction. Replace the separate
store.UpsertWorkbenchRevision, store.UpsertWorkbenchPrompt, and KV/prompt calls
with that atomic operation while preserving existing record contents and
local-cache updates only after persistence succeeds.
- Around line 1644-1704: Update organizationUUID initialization in
handleWorkbenchGeneratePrompt to use the registered orgUuid route parameter,
preferably via the existing workbenchOrgUUID helper, so all subsequent log
attributes contain the organization UUID.

---

Nitpick comments:
In `@internal/codesessions/ingress.go`:
- Around line 865-910: Factor the repeated request metadata from
logCodeSessionWorkerInternalEventsBadRequest and logCodeSessionWorkerOTLPRequest
into a shared helper or reusable field construction, including path extraction
and common request fields. Keep each logger’s event-specific fields—such as
signal, epoch details, reason, and error—unchanged while preserving the existing
log output.

In `@internal/codesessions/service.go`:
- Around line 54-62: Preserve context-derived fields by replacing the affected
injected logger calls with their context-aware variants, passing the available
request or worker context. Apply this to internal/codesessions/service.go lines
54-62, 97, and 359; internal/codesessions/tool_permissions.go lines 37-40;
internal/environments/runner.go line 78 using ctx; internal/files/platform.go
changed request-handler logs through line 250;
internal/sessions/code_event_bridge.go lines 18, 40, and 48; and
internal/sessions/service.go changed request-handler logs through line 792. Use
DebugContext, InfoContext, WarnContext, or ErrorContext matching each existing
log level.

In `@internal/files/platform_test.go`:
- Around line 26-33: Update the platform logging tests around capturePlatformLog
and the cases at the referenced failure paths to capture slog.Record values
instead of rendered log text. Assert each record’s level, message, and
structured attributes—including variant, file_uuid, key, bytes_copied, and
expected_size—while preserving the existing error scenarios and readErr details
where applicable.

In `@internal/skills/handler.go`:
- Line 163: Replace the changed Error calls with context-aware ErrorContext
calls, passing r.Context() in internal/skills/handler.go:163-163 and
internal/vaults/handler.go:194-194, and ctx in internal/skills/seed.go:107-107.
Preserve each existing log message and attributes.
- Around line 730-748: The object-storage logs use the ambiguous attribute name
key; rename it to object_key in the affected logs in internal/skills/handler.go
lines 730-748 and internal/skills/seed.go lines 123-126, including the stream,
open, and seed-related logging calls as applicable. No other log attributes or
behavior need changing.

In `@internal/workbench/workbench_handler.go`:
- Around line 38-44: The workbench persistence adapter silently falls back when
the store does not implement workbenchPersistenceStore. Update
workbenchPersistenceFromStore and its use in newWorkbenchHandler to detect a
failed type assertion and emit a warning through the injected logger, while
preserving the existing fallback behavior and normal path when the assertion
succeeds.