[FEAT] Allow unpublishing an exported tool; derive API-key target from the URL

[hari-kuriakose]

Purpose

Two API ergonomics fixes: unpublishing an exported tool, and creating an API key without repeating an identifier that is already in the URL. Review surfaced a same-org IDOR on the API-key path, which is closed here as well — see §2.

A third related item — making a no-RAG profile omit the vector DB / embedding model — is deliberately excluded; see "Descoped" below.

---

1. No way to delete an exported registry tool

The registry is read-only over the API — prompt_studio_registry_v2/urls.py maps only {"get": "list"}. The only way to remove an entry is to delete the Prompt Studio project, which cascades to it. That is implicit, undocumented, and blunt: a tool cannot be unpublished while keeping the project.

Adds DELETE registry/<pk>/, guarded by the same in-use check the Prompt Studio project delete already performs — a tool still attached to a workflow is refused with 409 rather than silently breaking those workflows. The guard filters ToolInstance on tool_id=instance.pk, matching the existing check in prompt_studio_core_v2/views.py:230 (_check_tool_usage_in_workflows; the check itself is at :244) — an exported tool's tool_id is its prompt_registry_id.

The 409 names the blocking deployment types (API Deployment / ETL Pipeline / Task Pipeline / Human in the Loop) by running the same _get_deployment_types logic as the core path (:249), so the refusal is actionable rather than a bare "no".

Unpublishing is not reversible in place. Re-exporting mints a fresh prompt_registry_id and does not carry over shared_to_org / shared_users. Nothing dangles — but anything holding the old UUID (saved workflow JSON, Postman collections, docs) silently stops resolving and must be updated.

Detail-route queryset: get_queryset previously returned None when no query-param filters were present, which would break get_object() on a detail route. It now returns the full queryset when addressing a single row by PK, keyed off self.kwargs.get("pk") rather than self.detail — DRF sets cls.detail = None under a manual as_view() (it is only populated for router-generated views), which this viewset uses. The existing list route is unaffected (no pk kwarg → same filter path as before).

Authorization — this is the first by-PK operation on this viewset (it previously exposed only list). DEFAULT_PERMISSION_CLASSES is empty and the viewset declared no permission_classes. OrganizationFilterBackend runs inside get_object() (via filter_queryset), so cross-org deletion was never possible — but any member of the same org could otherwise delete another member's exported tool by PK. IsRegistryToolOwner gates only destroy:

• Ownership inherited from the linked CustomTool, mirroring IsParentToolOwner (same pattern for ProfileManager).
• Falls back to the row's own owner for unlinked legacy rows (custom_tool is nullable).
• Service accounts and org admins admitted, matching sibling permission classes.
• list visibility is untouched — still derived by list_tools. Only the destructive route is restricted.

2. API-key creation requires an identifier already present in the URL

POST keys/api/<api_id>/ takes api_id as a path segment but also expects api in the body — the same value, spelled twice. Omitting the body field fails validation. POST keys/pipeline/<pipeline_id>/ has the identical shape.

POST routes to the default ModelViewSet.create, which never sees the URL kwargs, so the body has to repeat them. create now derives the target from the path when present, and falls through to the default implementation for the body-only routes (keys/api/, keys/pipeline/) — both keep working.

The path target is authoritative, not a default: a body naming the other target is a contradiction and is refused with a 400 rather than silently creating a key for whichever one wins. Assigning rather than setdefault-ing also means {"api": ""} no longer defeats the derivation into a confusing 400, and a non-mapping (JSON array) body is rejected as a 400 instead of AttributeError-ing into a 500.

Closes an IDOR on key creation

create performed no ownership check on the target deployment. get_permissions returns IsParentDeploymentOwner for it, but that class implements only has_object_permission — and create is collection-level, so DRF never calls get_object() / check_object_permissions. has_permission fell through to BasePermission's default True. Any authenticated org member could POST keys/api/<someone-else's-api-id>/ and mint a live key for a deployment they do not own. (Cross-org was already blocked by the org-scoped default manager; this was same-org.)

This pre-existed the PR — the body-based path had the identical gap — but the change makes it reachable from the path segment alone and touches exactly the method where the fix belongs, so it is closed here. create now resolves the target from path or body and object-checks it on every route: guarding only the path form would have left the identical hole reachable by moving the identifier into the body.

A malformed identifier is also handled at the fetch boundary. pk is a UUID column, so a non-UUID value raises ValidationError (not DoesNotExist) out of to_python; unguarded, reading the target from the body before the serializer runs would turn ordinary client garbage into a 500. get_api_by_id / get_pipeline_by_id now treat it as "not found".

That required a fix in IsParentDeploymentOwner itself. Its body reads obj.api or obj.pipeline or obj, and the target handed to it is an APIDeployment / Pipeline — neither of which declares an api or pipeline field (APIKey.api points at the deployment, related_name="api_keys"). A bare attribute access would raise AttributeError → 500 on every key creation. The lookups are now getattr guarded, falling through to obj; both parents carry memberships, so _is_resource_owner resolves correctly against them. Behaviour on APIKey detail routes is unchanged.

The pipeline branch uses a new PipelineProcessor.get_pipeline_by_id rather than get_active_pipeline. Minting a key does not require a running pipeline, and get_active_pipeline would both 422 on a paused one (Pipeline.active defaults to False) and disclose its state before the ownership check — the exact information the check exists to protect. get_active_pipeline itself is untouched.

---

Descoped: making a no-RAG profile omit the vector DB / embedding model

One candidate change was to make vector_store and embedding_model nullable so a chunk_size=0 ("no RAG") profile need not supply adapters that are never read. It is excluded because the premise does not hold in the code.

The rationale would be that with chunk_size=0 neither the vector DB nor the embedding model is queried. But build_single_pass_payload sets default_profile.chunk_size = 0 at prompt_studio_helper.py:1140 and then reads both FKs 25 lines later, in the same function:

default_profile.chunk_size = 0        # :1140
...
vector_db = str(default_profile.vector_store.id)        # :1165
embedding_model = str(default_profile.embedding_model.id)  # :1166

Export does the same unconditionally (prompt_studio_registry_helper.py:269-270). Across the backend there are 46 unguarded reads of these two FKs, spanning indexing, export, single-pass, and permission validation.

Making the columns nullable without guarding every reader would convert a clear 400 at profile creation into AttributeError: 'NoneType' object has no attribute 'id' — a 500 deep inside single-pass or export, in exactly the mode the change claims is safe. The serializer-only alternative does not escape this: storing null still requires the migration, which exposes the same readers.

There is no ergonomics-sized version of this change. It needs the migration plus a deliberate decision about what each read path does with an absent adapter — its own PR.

---

Verification

• ruff and ruff-format pass at the version pinned in .pre-commit-config.yaml (v0.3.4).
• No model or migration changes in this PR.
• 41 unit tests pass (api_v2/tests/test_api_key_create_target.py, prompt_studio/prompt_studio_registry_v2/tests/test_registry_tool_delete_guards.py), under both cd backend && pytest and repo-root pytest backend/....
• Tests execute the real method bodies rather than asserting on source text, and are verified by mutation. Each of these fails the suite: re-opening the body-only IDOR, making the authorization check fail open, unwiring destroy from its guard, narrowing or broadening the malformed-id catch, dropping a deployment-type branch, and breaking the tool_id filter.
• Django settings are not configured in the unit tier, so route binding and the live permission cycle (which need a database) remain integration-tier; backend/conftest.py auto-marks such tests integration. The guard logic is covered per-PR.
• Pre-existing collection errors in the wider backend suite (9 modules, ImproperlyConfigured) are unchanged by this PR — verified against the base commit.

Review

Went through the team's standardized 15-lens review. Three rounds; each found real defects in the preceding round's fix, which are folded into the commits above:

1. The path-only fix left the body-only route open (Critical), the pipeline lookup narrowed a status contract and leaked state (High), the authorization check failed open on unexpected types (High), and two test suites passed against broken code (High).
2. Widening create to read the body introduced a 500 on malformed identifiers (High).
3. The helper tests for that fix asserted on source text rather than behaviour (Medium).

Round 4 is clean. A separate behaviour-preserving commit shares the test extraction helper; the reviewed source files are byte-identical across it.

Impact

• A tool can be unpublished without deleting its Prompt Studio project; in-use tools are refused with 409, and deletion is restricted to the project's owner / org admins.
• The path identifier alone is enough to create an API key; existing body-based callers are unaffected.
• API-key creation now verifies ownership of the target deployment/pipeline, closing a same-org IDOR that predates this PR.
• A body contradicting the path target is refused with a 400 instead of resolving to one of them.

Related

Part of a set of independent Prompt Studio / registry fixes: #2203, #2204, #2209. No code dependency between them.

🤖 Generated with Claude Code

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

Walkthrough

Adds nested-route API key creation and introduces guarded registry deletion with owner authorization, UUID detail routing, query-safe object lookup, and a 409 conflict when workflows still reference a registry tool.

Changes

Nested API key creation

Layer / File(s)	Summary
Path-derived API key creation backend/api_v2/api_key_views.py	APIKeyViewSet.create populates missing API or pipeline fields from URL parameters, validates and creates the key, and retains the superclass flow for unscoped requests.

Guarded registry deletion

Layer / File(s)	Summary
Registry deletion route and authorization backend/prompt_studio/prompt_studio_registry_v2/urls.py, backend/prompt_studio/prompt_studio_registry_v2/views.py, backend/prompt_studio/permission.py, backend/prompt_studio/prompt_studio_registry_v2/tests/test_registry_tool_delete_guards.py	A UUID detail route maps DELETE requests to destroy; routed object lookup bypasses query-parameter filtering, and owner authorization covers linked, legacy, administrator, and service-account cases.
Workflow reference deletion guard backend/prompt_studio/prompt_studio_registry_v2/views.py, backend/prompt_studio/prompt_studio_registry_v2/exceptions.py, backend/prompt_studio/prompt_studio_registry_v2/tests/test_registry_tool_delete_guards.py	Registry deletion checks distinct dependent workflow IDs, raises a 409 RegistryToolInUseError when references exist, and otherwise delegates to the base destroy handler.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  actor Client
  participant PromptStudioRegistryView
  participant IsRegistryToolOwner
  participant ToolInstance
  Client->>PromptStudioRegistryView: DELETE registry/{pk}
  PromptStudioRegistryView->>IsRegistryToolOwner: Check destroy permission
  IsRegistryToolOwner-->>PromptStudioRegistryView: Allow or deny request
  PromptStudioRegistryView->>ToolInstance: Query workflow references
  ToolInstance-->>PromptStudioRegistryView: Return distinct workflow IDs
  PromptStudioRegistryView-->>Client: 409 conflict or deletion response

Loading

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 63.53% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly summarizes both primary changes: registry tool unpublishing and URL-derived API-key targets.
Description check	✅ Passed	The description thoroughly explains the changes, rationale, risks, testing, migrations, configuration, and related issues.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/prompt-studio-ergonomics

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Greptile Summary

The PR adds guarded registry-tool unpublishing and derives API-key targets from request URLs while enforcing target ownership.

• Adds a DELETE route for exported registry tools with owner/admin authorization and workflow-usage refusal.
• Consolidates workflow dependency and deployment-type reporting helpers.
• Derives API or pipeline targets during API-key creation, rejects contradictory inputs, and checks object permissions across path- and body-based routes.
• Handles malformed deployment identifiers as not found and adds focused behavioral and route-wiring tests.

Confidence Score: 5/5

The PR appears safe to merge because no additional blocking failure eligible for this follow-up review remains.

No blocking failure remains beyond the dependency check-delete race already documented in the existing inline thread.

Important Files Changed

Filename	Overview
backend/prompt_studio/prompt_studio_registry_v2/views.py	Adds the organization-scoped, owner-gated registry deletion route and refuses deletion when existing workflow references are found.
backend/prompt_studio/tool_usage.py	Centralizes workflow dependency lookup and deployment-type formatting for registry and Prompt Studio deletion paths.
backend/prompt_studio/permission.py	Adds object-level ownership authorization for unpublishing registry tools.
backend/api_v2/api_key_views.py	Resolves API-key targets from either URL or body, rejects contradictory targets, and explicitly applies object permissions before creation.
backend/permissions/permission.py	Extends parent-deployment authorization to safely recognize both API-key rows and their parent resources.
backend/pipeline_v2/pipeline_processor.py	Adds inactive-agnostic pipeline lookup and converts malformed UUID lookup errors into not-found results.
backend/api_v2/utils.py	Converts malformed API deployment identifiers into not-found results.
backend/tests_common/test_route_wiring.py	Verifies that the new guarded methods are bound to their intended routes and permission classes.

_{Reviews (10): Last reviewed commit: "[FIX] Unbreak pre-commit.ci: attribute d..." | Re-trigger Greptile}

[coderabbitai]

Actionable comments posted: 3

🧹 Nitpick comments (1)

backend/prompt_studio/prompt_studio_registry_v2/views.py (1)

50-52: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Correct the variadic parameter annotations.

*args: tuple[Any] annotates each positional argument as a tuple, while **kwargs: dict[str, Any] annotates each keyword value as a dictionary; type checkers interpret variable-argument annotations this way. Use *args: Any, **kwargs: Any (or an existing project-standard DRF-compatible signature).

Proposed fix

     def Destroy(
-        self, request: Request, *args: tuple[Any], **kwargs: dict[str, Any]
+        self, request: Request, *args: Any, **kwargs: Any
     ) -> Response:

🤖 Prompt for AI Agents

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

In `@backend/prompt_studio/prompt_studio_registry_v2/views.py` around lines 50 -
52, Update the destroy method signature to annotate variadic parameters as
individual values: use Any for both args and kwargs rather than tuple[Any] and
dict[str, Any]. Preserve the existing return type and method behavior.

🤖 Prompt for all review comments with AI agents

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 `@backend/api_v2/api_key_views.py`:
- Around line 40-43: Update the request-data target injection logic in the API
key view so the URL-derived api_id or pipeline_id is added only when neither
“api” nor “pipeline” is already present in the body. Preserve body values and
prevent both target fields from being populated before APIKeySerializer.validate
runs.

In `@backend/prompt_studio/prompt_studio_registry_v2/views.py`:
- Around line 61-69: Update the dependency guard in the tool deletion flow to
use a queryset `.exists()` check instead of materializing workflow IDs with
`set(values_list(...).distinct())`. Keep the existing deletion prevention
behavior, and change the `logger.info` call to report only a count or generic
dependency message rather than listing workflow IDs.
- Around line 60-72: Update the destroy flow containing
PromptStudioRegistryViewSet.destroy so the dependent-workflow check and registry
deletion execute within one database transaction, locking the registry and
reusing that locked instance rather than allowing super().destroy() to re-fetch
it. Add database-level protection or an equivalent transactional safeguard for
concurrent ToolInstance attachments, preserving RegistryToolInUseError when
dependencies exist, and add a concurrency test covering an attachment racing
with deletion.

---

Nitpick comments:
In `@backend/prompt_studio/prompt_studio_registry_v2/views.py`:
- Around line 50-52: Update the destroy method signature to annotate variadic
parameters as individual values: use Any for both args and kwargs rather than
tuple[Any] and dict[str, Any]. Preserve the existing return type and method
behavior.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: ac14b42f-0572-4c09-a8dd-a6eaf024961a

📥 Commits

Reviewing files that changed from the base of the PR and between 023b140 and a26122b.

📒 Files selected for processing (4)

• backend/api_v2/api_key_views.py
• backend/prompt_studio/prompt_studio_registry_v2/exceptions.py
• backend/prompt_studio/prompt_studio_registry_v2/urls.py
• backend/prompt_studio/prompt_studio_registry_v2/views.py

[coderabbitai]

Actionable comments posted: 19

Note

Due to the large number of review comments, Critical, Major severity comments were prioritized as inline comments.

🟡 Minor comments (6)

prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/base.py-46-53 (1)

46-53: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Validate output before indexing.

output = payload.get(PSKeys.OUTPUT) yields None when the key is absent, and lines 51-53 immediately index into it (output[PSKeys.NAME]), producing an unhandled TypeError/500 instead of a clear BadRequest.

🛡️ Proposed fix

     output = payload.get(PSKeys.OUTPUT)
     tool_id: str = payload.get(PSKeys.TOOL_ID, "")
     file_hash = payload.get(PSKeys.FILE_HASH)
     structured_output: dict[str, Any] = {}
+    if not output:
+        raise BadRequest("No output provided in the request.")

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/base.py`
around lines 46 - 53, Validate that output retrieved in the payload-processing
method is present before accessing PSKeys.NAME or PSKeys.PROMPT. If it is
missing or invalid, raise the established BadRequest error with a clear message;
otherwise preserve the existing variable_names, prompt_name, and promptx
initialization flow.

prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/README.md-5-5 (1)

5-5: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Document the registered endpoint.

The README advertises /answer-sps, but SimplePromptStudio registers /answer-prompt-public in prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/base.py. Update the README to prevent clients from receiving 404 responses.

Proposed fix

-If the plugin is not disabled, it registers a new endpoint `/answer-sps` which extracts the prompt result without authentication.
+If the plugin is not disabled, it registers a new endpoint `/answer-prompt-public` which extracts the prompt result without authentication.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/README.md`
at line 5, Update the SimplePromptStudio README to document the registered
endpoint as /answer-prompt-public instead of /answer-sps, matching the route
defined by the SimplePromptStudio implementation and preventing clients from
using the obsolete path.

prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/interface/runner.py-199-204 (1)

199-204: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

headers in the final output reflects only the last processed sheet. For multi-sheet workbooks, headers is the loop variable left over from the final iteration, so metadata.detected_headers misrepresents earlier sheets. Consider aggregating headers per sheet (or dropping the single headers field from the assembled metadata).

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/interface/runner.py`
around lines 199 - 204, Update the SmartTableExtractorRunner final-output
assembly so metadata does not use the loop-scoped headers from only the last
processed sheet. Aggregate detected headers for every sheet and pass that
collection to _assemble_final_output, or remove the single headers field from
the assembled metadata while preserving accurate multi-sheet output.

prompt-service/src/unstract/prompt_service/plugins/evaluation/src/base.py-172-176 (1)

172-176: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Missing eval_settings keys are reported as failures, not "disabled".

self.settings defaults to {} (Line 88), so self.settings["evaluate"] here (and self.settings["monitor_llm"] at Line 180) raise KeyError when a prompt has no eval settings. That KeyError is caught by the outer handler at Line 242 and re-raised as EvalFailedError, whereas the intended semantics for an unconfigured prompt is EvalDisabledError. Prefer .get(...) with an explicit disabled check.

🛠️ Proposed fix

-            if not self.response or self.settings["evaluate"] is not True:
+            if not self.response or self.settings.get("evaluate") is not True:
                 raise EvalDisabledError()

🤖 Prompt for AI Agents

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

In `@prompt-service/src/unstract/prompt_service/plugins/evaluation/src/base.py`
around lines 172 - 176, Update the evaluation checks in the relevant base-class
flow around the response fallback and the `monitor_llm` handling to use
`self.settings.get(...)` with explicit disabled defaults instead of direct key
access. Ensure prompts with missing `evaluate` or `monitor_llm` settings raise
`EvalDisabledError` rather than allowing `KeyError` to become `EvalFailedError`,
while preserving enabled-setting behavior.

prompt-service/src/unstract/prompt_service/plugins/challenge/src/base.py-101-114 (1)

101-114: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

LLM failure silently returns a passing score.

If challenge_llm.complete raises (caught at Line 107), current_answer is never bound. Line 110 then hits UnboundLocalError, which is caught by the broad except Exception at Line 111, logged as a misleading "JSON format error", and the method returns default_answer (score: 5). A genuine LLM/transport failure is thus indistinguishable from a valid pass and skips the retry path in run().

🐛 Proposed fix

         try:
             completion = self.challenge_llm.complete(
                 prompt=prompt,
             )
             current_answer = completion["response"].text
-        # TODO: Use another LLM to complete the prompt
         except Exception as e:
             app.logger.error("Error completing prompt: %s.", str(e))
+            return default_answer
         try:
             return json.loads(current_answer)

🤖 Prompt for AI Agents

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

In `@prompt-service/src/unstract/prompt_service/plugins/challenge/src/base.py`
around lines 101 - 114, Update the completion and parsing flow in the method
containing challenge_llm.complete so an LLM failure does not fall through to
JSON parsing or return default_answer. Handle the exception by propagating or
returning a failure result that run() can recognize and retry, while reserving
the “JSON format error” path for responses that were actually received but
cannot be parsed.

prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/postprocessing.txt-14-19 (1)

14-19: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Malformed JSON in the example schema — missing closing quote.

Line 16's "line_no_end": "0x1, is missing the closing " before the comma (compare with line 15's correctly-formed "line_no_start": "0x1",). This is the reference schema shown to the LLM for generating the post-processing script; a malformed example risks confusing field/type inference.

🐛 Proposed fix

-        "line_no_end": "0x1,
+        "line_no_end": "0x1",

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/postprocessing.txt`
around lines 14 - 19, The example schema in the post-processing prompt contains
malformed JSON: fix the line_no_end field value in the schema example by adding
its missing closing quote, matching the valid line_no_start representation and
preserving the intended string type.

🧹 Nitpick comments (12)

prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/constants.py (1)

43-46: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Duplicate FILE_PATH definition.

FILE_PATH = "file_path" is declared on both line 43 and line 46; the second is redundant. Harmless (same value) but worth removing to avoid confusion.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/constants.py`
around lines 43 - 46, Remove the redundant second FILE_PATH constant declaration
in the constants module, preserving the first FILE_PATH = "file_path" definition
and all other constants unchanged.

prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/__init__.py (1)

6-6: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Sort the public export list.

Ruff reports RUF022 here. Use ["BatchProcessor", "HeaderDetector"] to clear the lint finding.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/__init__.py`
at line 6, Sort the __all__ export list alphabetically by placing BatchProcessor
before HeaderDetector to resolve Ruff RUF022.

Source: Linters/SAST tools

prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/batch_processor.py (1)

98-100: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Preserve the traceback in error logs.

Use logger.exception("Failed to process batches") inside this handler; SonarCloud currently flags the missing exception traceback.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/batch_processor.py`
around lines 98 - 100, Update the exception handler in the batch-processing flow
to use logger.exception("Failed to process batches") instead of logger.error,
preserving the active traceback while keeping the existing
BatchProcessingException propagation unchanged.

Source: Linters/SAST tools

prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/converter/excel_to_tsv.py (1)

128-146: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Use logger.exception(...) inside these except blocks. These handlers log with logger.error while an exception is in flight; logger.exception captures the traceback and clears the SonarCloud failures reported on lines 129/135/143 (also applies to the except Exception at line 152/153).

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/converter/excel_to_tsv.py`
around lines 128 - 146, The exception handlers in the Excel-to-TSV conversion
flow should use logger.exception instead of logger.error so active exception
tracebacks are captured. Update the handlers for
requests.exceptions.ConnectionError, requests.exceptions.RequestException, and
the broad Exception around the conversion logic, preserving their existing
messages and FileConversionException behavior.

Source: Linters/SAST tools

prompt-service/src/unstract/prompt_service/plugins/evaluation/src/__init__.py (1)

13-13: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Global ResourceWarning suppression leaks beyond this plugin.

warnings.simplefilter mutates the process-wide filter at import time, so any host that imports this plugin loses ResourceWarning visibility everywhere (potentially masking unrelated leaks). Consider scoping it with warnings.catch_warnings()/filterwarnings around the evaluator execution instead of a module-import side effect.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/evaluation/src/__init__.py`
at line 13, Remove the module-import-time warnings.simplefilter call in the
plugin initializer and scope the ResourceWarning suppression to the evaluator
execution path instead. Use warnings.catch_warnings with filterwarnings around
the relevant evaluation operation, ensuring the process-wide warning filter is
restored after execution.

prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py (3)

114-118: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Re-raise without exception chaining.

raise PluginException(...) inside the except FileNotFoundError block loses the original traceback context.

♻️ Proposed fix

         except FileNotFoundError as file_not_found:
             raise PluginException(
                 f"Input file {self.input_file} is "
                 f"not found in the path : {file_not_found}"
-            )
+            ) from file_not_found

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py`
around lines 114 - 118, Update the FileNotFoundError handler in the extractor’s
exception flow to re-raise PluginException without chaining the original
exception, while preserving the existing message content. Use explicit exception
suppression on the raise in the except FileNotFoundError block.

Source: Linters/SAST tools

---

421-442: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Large blocks of commented-out code left in place.

These dead-code blocks (header-normalization logic and header-prepending logic) add noise without being active. Consider removing or moving to a comment/ADR if kept for reference.

Also applies to: 549-553

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py`
around lines 421 - 442, Remove the commented-out header normalization and
header-prepending blocks near the affected sections, including the corresponding
block around the later referenced lines. Keep only active implementation code;
do not retain dead code inline, and preserve any required behavior through the
existing live logic.

---

54-54: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Mutable default arguments (headers: list = []) repeated across 5 signatures.

Flagged by Ruff (B006) and SonarCloud. None of these currently mutate the default in place, so there's no active aliasing bug, but it's a latent footgun if future edits mutate the list directly.

♻️ Example fix pattern

-    def extract_header(self, page_no: int, headers: list[str] = []) -> list[str]:
+    def extract_header(self, page_no: int, headers: list[str] | None = None) -> list[str]:
+        headers = headers or []

Also applies to: 217-217, 270-270, 326-326, 481-481

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py`
at line 54, Replace the mutable [] defaults for headers in all five affected
function signatures with a non-shared default, such as None, and initialize an
empty list inside each function when needed. Preserve existing behavior for
callers that provide headers, and update every headers signature including those
near the referenced locations.

Source: Linters/SAST tools

prompt-service/src/unstract/prompt_service/plugins/table_extractor/pyproject.toml (1)

6-6: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

flake8 listed as a runtime dependency.

flake8 is a linting tool and isn't imported anywhere in this plugin's source; it appears to be a dev-only tool accidentally added to dependencies instead of a dev/lint extras group.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/pyproject.toml`
at line 6, Remove flake8 from the runtime dependencies in the pyproject.toml
dependencies declaration; keep only the packages required by the plugin at
runtime, and do not add a replacement dev group unless one already exists for
linting.

prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/base_processor.py (1)

82-82: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

TODO left in shipped code.

# TODO Extract document metadata : indicates unfinished work. Happy to help implement metadata extraction here if desired.

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/base_processor.py`
at line 82, Remove the unfinished “Extract document metadata” TODO from the
affected processor code, or implement the metadata extraction before retaining
any note; do not leave the TODO in shipped code.

prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/post_processor.py (2)

27-27: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Use logger.exception() in exception handlers; CI (SonarCloud) is failing on these lines.

The SonarCloud check flags lines 66, 144-145, and 154 to use logging.exception() (which captures the stack trace) instead of logger.error(f"... {e}"). run_python_program_on_response (line 27) is also flagged for cognitive complexity exceeding the configured threshold (17 vs 15).

Also applies to: 66-66, 144-145, 153-154

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/post_processor.py`
at line 27, Update exception handlers in run_python_program_on_response and the
handlers around the flagged lines to use logger.exception() instead of
logger.error(f"... {e}"), preserving the existing context messages while
allowing stack traces to be captured. Reduce run_python_program_on_response’s
cognitive complexity below the configured threshold by extracting cohesive logic
into small helper functions without changing behavior.

Source: Pipeline failures

---

157-165: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win

Redundant double CSV parse; consider DataFrame.to_numpy().

coloumns is derived from a first pd.read_csv call and then immediately passed back as usecols to a second, identical parse of the same TSV — the second read only re-parses the same data with no filtering effect. Also, df.values.tolist() is flagged by SonarCloud in favor of df.to_numpy().tolist().

♻️ Proposed simplification

     def _process_csv_to_json(cleaned_tsv: str) -> dict[str, list[Any]]:
-        coloumns = pd.read_csv(StringIO(cleaned_tsv), sep="\t").columns
-        df = pd.read_csv(StringIO(cleaned_tsv), sep="\t", usecols=coloumns)
+        df = pd.read_csv(StringIO(cleaned_tsv), sep="\t")
         df.fillna("", inplace=True)
         output: dict[str, list[Any]] = {
             "column_headers": df.columns.tolist(),
-            "rows": df.values.tolist(),
+            "rows": df.to_numpy().tolist(),
         }
         return output

🤖 Prompt for AI Agents

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

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/post_processor.py`
around lines 157 - 165, Update _process_csv_to_json to parse the TSV only once,
removing the redundant coloumns read and usecols filtering, then preserve the
existing fillna and output structure. Replace df.values.tolist() with
df.to_numpy().tolist() when building rows.

🤖 Prompt for all review comments with AI agents

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
`@prompt-service/src/unstract/prompt_service/plugins/evaluation/src/frameworks/base.py`:
- Around line 190-275: Reduce cognitive complexity in gen_final_results by
extracting key parsing, judge-remark aggregation, and post-processing into
focused helper methods. Preserve the existing score averaging, quorum
validation, feedback construction, and _format_result behavior, including
score_total, line-item ID/name, release, and disabled handling. Keep
gen_final_results as the orchestration method that iterates judges_remarks and
appends each formatted result.
- Around line 277-305: The run method returns unresolved evaluator futures,
allowing callers to read gen_final_results before callbacks update shared state.
Update run and its caller flow to await or resolve every item in the returned
futures collection before invoking gen_final_results or consuming evaluator
output, while preserving the existing evaluator scheduling behavior.

In
`@prompt-service/src/unstract/prompt_service/plugins/line_item_extraction/src/base.py`:
- Around line 46-47: Convert the LINE_ITEM_EXTRACTION_MAX_LLM_CALLS environment
value to an integer when assigning MAX_ATTEMPTS, while retaining 5 as the
integer default, so the attempts comparison in the loop remains type-compatible.

In
`@prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/base.py`:
- Around line 161-173: In the challenge setup flow, initialize challenge_llm to
None before plugin lookup, make exception handling catch a concrete exception
type without subscripting challenge_plugin, and build challenge_metrics only
when challenge_llm was successfully created rather than whenever
enable_challenge is truthy. Preserve the existing no-plugin logging and metrics
behavior for successful challenge initialization.

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/interface/runner.py`:
- Around line 240-242: In the JSONDecodeError handler within the schema parsing
flow, update the SmartTableExtractorKeys.SCHEMA assignment to use the repaired
value returned by repair_json(schema) rather than the original invalid schema.
Ensure the repaired variable is consumed so downstream processing receives the
corrected schema and no unused-variable warning remains.
- Around line 74-104: Update SmartTableExtractorRunner.run to obtain the schema
and input_file from table_settings when the corresponding arguments are not
provided, so _validate_inputs and file processing receive valid values. Default
a missing fs_instance to the local filesystem provider before calling read,
while preserving explicitly supplied providers for remote files. Add the
required provider import and keep the existing extraction flow unchanged.

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/batch_processor.py`:
- Around line 114-124: Update header detection to return both the detected
headers and their source row index, rather than inferring the index from column
count alone. Propagate this index into batching and slice rows starting
immediately after that exact header position, ensuring same-width title or
preamble rows are not treated as headers.

In `@prompt-service/src/unstract/prompt_service/plugins/summarize/src/base.py`:
- Around line 31-45: Update the payload validation used by the summarize request
flow around validate_payload and enhance_summarize_prompt to enforce that the
JSON body is an object and optional fields have their expected types, especially
ensuring prompt_keys is a list before prompt construction. Reject invalid field
types with the existing BadRequest path so malformed requests do not reach
enhance_summarize_prompt or produce a 500 response.
- Around line 60-64: Update the exception handler in the summarize method to
raise the constructed InternalServerError(error) instead of merely instantiating
it, ensuring LLM failures do not fall through to return result or produce a
successful response. Preserve the existing logging and error message.

In `@prompt-service/src/unstract/prompt_service/plugins/summarize/src/helper.py`:
- Around line 15-18: Validate the request payload before the prompt-building
logic around prompt_keys and before calling Summarize.summarize: require a JSON
object, ensure prompt_keys is a list containing only strings, and reject null or
blank required fields. Return HTTP 400 for any malformed payload, and only
execute the existing prompt construction when validation succeeds.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py`:
- Around line 120-146: Avoid mutating the shared prompt definitions in
__initalize_prompts by deep-copying Prompts.BASE_PROMPTS[mode] before resolving
file references and updating self.prompts. In
prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py
lines 120-146, apply the change at the assignment before the existing loop;
prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/prompts.py
lines 5-33 requires no direct change and should remain a read-only shared
structure.
- Around line 493-535: Make the continuation-page prompt construction in the
loop following the initial extraction use the same column count as the
first-page prompt: replace the plain column_count substitution with the value
including the two page/line-number columns. Keep the existing headers and
downstream parsing flow unchanged.
- Around line 180-214: Initialize the local headers variable to an empty list
before the response branch in extract_header. Preserve the existing
parsed-header assignment for non-empty, non-"na" responses, and ensure the
empty/"na" path returns the initialized empty list without raising
UnboundLocalError.
- Around line 587-628: Update process_raw_jsonl_to_tsv so each row is parsed
once, retain the parsed object through validation and output, and add x_page
before establishing or using headers so the page value is included in
cleaned_tsv. Replace both broad silent exception handlers with targeted JSON
parsing/error handling that logs malformed-row failures before continuing.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/base_processor.py`:
- Line 26: The document type fallback is resolved inconsistently between
TableExtractionBase and the runner. In
prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/base_processor.py#L26,
stop independently deriving the value or use runner.py’s "default" fallback; in
prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/interface/runner.py#L30,
resolve document_type once in run_table_extraction and pass it to
TableExtractionBase.extract_large_table so both consumers use the same value.
- Around line 82-101: Update the empty-header handling around
large_table_extractor.extract_header in the processor flow so that when
document_type is absent or does not match bank_statement or rent_rolls, the code
explicitly logs the missing headers and raises the appropriate PluginException
instead of falling through with an empty list. Preserve the existing
default-header behavior for bank_statement and the existing failure behavior for
rent_rolls.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/post_processor.py`:
- Line 20: Change the TABLE_DEBUG default in the post-processing configuration
to false so temporary generated scripts and extracted TSV data are deleted by
the existing finally cleanup unless debugging is explicitly enabled with
TABLE_DEBUG=true.
- Around line 119-148: Update _run_processing_script so all environments execute
the untrusted script through the sandboxed Docker command with the existing
network, memory, CPU, and nonroot restrictions; do not run script_path directly
based on is_k8s or is_docker. Add a finite timeout to subprocess.run and handle
subprocess.TimeoutExpired explicitly, preserving the existing
PluginException-based failure reporting.
- Around line 149-155: Update the exception handlers in the post-processing
function to propagate the original failure after logging, including both
FileNotFoundError and unexpected Exception cases, instead of falling through
with an implicit None return. Preserve the existing timeout handling unless
needed for consistent error propagation, and ensure the caller receives the
actual underlying error rather than failing later at output.strip().

---

Minor comments:
In `@prompt-service/src/unstract/prompt_service/plugins/challenge/src/base.py`:
- Around line 101-114: Update the completion and parsing flow in the method
containing challenge_llm.complete so an LLM failure does not fall through to
JSON parsing or return default_answer. Handle the exception by propagating or
returning a failure result that run() can recognize and retry, while reserving
the “JSON format error” path for responses that were actually received but
cannot be parsed.

In `@prompt-service/src/unstract/prompt_service/plugins/evaluation/src/base.py`:
- Around line 172-176: Update the evaluation checks in the relevant base-class
flow around the response fallback and the `monitor_llm` handling to use
`self.settings.get(...)` with explicit disabled defaults instead of direct key
access. Ensure prompts with missing `evaluate` or `monitor_llm` settings raise
`EvalDisabledError` rather than allowing `KeyError` to become `EvalFailedError`,
while preserving enabled-setting behavior.

In
`@prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/README.md`:
- Line 5: Update the SimplePromptStudio README to document the registered
endpoint as /answer-prompt-public instead of /answer-sps, matching the route
defined by the SimplePromptStudio implementation and preventing clients from
using the obsolete path.

In
`@prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/base.py`:
- Around line 46-53: Validate that output retrieved in the payload-processing
method is present before accessing PSKeys.NAME or PSKeys.PROMPT. If it is
missing or invalid, raise the established BadRequest error with a clear message;
otherwise preserve the existing variable_names, prompt_name, and promptx
initialization flow.

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/interface/runner.py`:
- Around line 199-204: Update the SmartTableExtractorRunner final-output
assembly so metadata does not use the loop-scoped headers from only the last
processed sheet. Aggregate detected headers for every sheet and pass that
collection to _assemble_final_output, or remove the single headers field from
the assembled metadata while preserving accurate multi-sheet output.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/postprocessing.txt`:
- Around line 14-19: The example schema in the post-processing prompt contains
malformed JSON: fix the line_no_end field value in the schema example by adding
its missing closing quote, matching the valid line_no_start representation and
preserving the intended string type.

---

Nitpick comments:
In
`@prompt-service/src/unstract/prompt_service/plugins/evaluation/src/__init__.py`:
- Line 13: Remove the module-import-time warnings.simplefilter call in the
plugin initializer and scope the ResourceWarning suppression to the evaluator
execution path instead. Use warnings.catch_warnings with filterwarnings around
the relevant evaluation operation, ensuring the process-wide warning filter is
restored after execution.

In
`@prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/constants.py`:
- Around line 43-46: Remove the redundant second FILE_PATH constant declaration
in the constants module, preserving the first FILE_PATH = "file_path" definition
and all other constants unchanged.

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/converter/excel_to_tsv.py`:
- Around line 128-146: The exception handlers in the Excel-to-TSV conversion
flow should use logger.exception instead of logger.error so active exception
tracebacks are captured. Update the handlers for
requests.exceptions.ConnectionError, requests.exceptions.RequestException, and
the broad Exception around the conversion logic, preserving their existing
messages and FileConversionException behavior.

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/__init__.py`:
- Line 6: Sort the __all__ export list alphabetically by placing BatchProcessor
before HeaderDetector to resolve Ruff RUF022.

In
`@prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/batch_processor.py`:
- Around line 98-100: Update the exception handler in the batch-processing flow
to use logger.exception("Failed to process batches") instead of logger.error,
preserving the active traceback while keeping the existing
BatchProcessingException propagation unchanged.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/pyproject.toml`:
- Line 6: Remove flake8 from the runtime dependencies in the pyproject.toml
dependencies declaration; keep only the packages required by the plugin at
runtime, and do not add a replacement dev group unless one already exists for
linting.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py`:
- Around line 114-118: Update the FileNotFoundError handler in the extractor’s
exception flow to re-raise PluginException without chaining the original
exception, while preserving the existing message content. Use explicit exception
suppression on the raise in the except FileNotFoundError block.
- Around line 421-442: Remove the commented-out header normalization and
header-prepending blocks near the affected sections, including the corresponding
block around the later referenced lines. Keep only active implementation code;
do not retain dead code inline, and preserve any required behavior through the
existing live logic.
- Line 54: Replace the mutable [] defaults for headers in all five affected
function signatures with a non-shared default, such as None, and initialize an
empty list inside each function when needed. Preserve existing behavior for
callers that provide headers, and update every headers signature including those
near the referenced locations.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/base_processor.py`:
- Line 82: Remove the unfinished “Extract document metadata” TODO from the
affected processor code, or implement the metadata extraction before retaining
any note; do not leave the TODO in shipped code.

In
`@prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/post_processor.py`:
- Line 27: Update exception handlers in run_python_program_on_response and the
handlers around the flagged lines to use logger.exception() instead of
logger.error(f"... {e}"), preserving the existing context messages while
allowing stack traces to be captured. Reduce run_python_program_on_response’s
cognitive complexity below the configured threshold by extracting cohesive logic
into small helper functions without changing behavior.
- Around line 157-165: Update _process_csv_to_json to parse the TSV only once,
removing the redundant coloumns read and usecols filtering, then preserve the
existing fillna and output structure. Replace df.values.tolist() with
df.to_numpy().tolist() when building rows.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: f41cb416-e5bd-4b6a-ae2b-671817aedc02

📥 Commits

Reviewing files that changed from the base of the PR and between 3519b75 and 8235086.

⛔ Files ignored due to path filters (9)

• prompt-service/src/unstract/prompt_service/plugins/challenge/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/evaluation/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/highlight_data/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/line_item_extraction/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/summarize/uv.lock is excluded by !**/*.lock
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/uv.lock is excluded by !**/*.lock

📒 Files selected for processing (85)

• backend/prompt_studio/permission.py
• backend/prompt_studio/prompt_studio_registry_v2/views.py
• prompt-service/src/unstract/prompt_service/plugins/challenge/README.md
• prompt-service/src/unstract/prompt_service/plugins/challenge/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/challenge/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/challenge/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/README.md
• prompt-service/src/unstract/prompt_service/plugins/evaluation/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/constants.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/frameworks/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/frameworks/base.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/frameworks/llama_index.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/frameworks/ragas.py
• prompt-service/src/unstract/prompt_service/plugins/evaluation/src/frameworks/unstract.py
• prompt-service/src/unstract/prompt_service/plugins/highlight_data/README.md
• prompt-service/src/unstract/prompt_service/plugins/highlight_data/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/highlight_data/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/highlight_data/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/highlight_data/src/constants.py
• prompt-service/src/unstract/prompt_service/plugins/line_item_extraction/README.md
• prompt-service/src/unstract/prompt_service/plugins/line_item_extraction/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/line_item_extraction/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/line_item_extraction/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/README.md
• prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/simple_prompt_studio/src/helper.py
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/README.md
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/constants.py
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/exceptions.py
• prompt-service/src/unstract/prompt_service/plugins/single_pass_extraction/src/helper.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/README.md
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/constants.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/converter/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/converter/excel_to_tsv.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/exceptions.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/interface/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/interface/runner.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/batch_processor.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/processor/header_detector.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/prompts/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/src/prompts/prompts.py
• prompt-service/src/unstract/prompt_service/plugins/smart_table_extractor/test_runner.py
• prompt-service/src/unstract/prompt_service/plugins/summarize/README.md
• prompt-service/src/unstract/prompt_service/plugins/summarize/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/summarize/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/summarize/src/base.py
• prompt-service/src/unstract/prompt_service/plugins/summarize/src/constants.py
• prompt-service/src/unstract/prompt_service/plugins/summarize/src/helper.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/README.md
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/pyproject.toml
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/constants.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/exceptions.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/extractor/core.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/interface/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/interface/helper.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/interface/runner.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/base_processor.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/processor/post_processor.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/__init__.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/extraction.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/headers.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/postprocessing.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/table_detect.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/table_span_contiguous.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/bank_statement/table_span_headers.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/prompts.py
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/rent_rolls/extraction.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/rent_rolls/headers.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/rent_rolls/postprocessing.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/rent_rolls/table_detect.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/rent_rolls/table_span_contiguous.txt
• prompt-service/src/unstract/prompt_service/plugins/table_extractor/src/prompts/rent_rolls/table_span_headers.txt

🚧 Files skipped from review as they are similar to previous changes (1)

• backend/prompt_studio/prompt_studio_registry_v2/views.py

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

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
`@backend/prompt_studio/prompt_studio_registry_v2/tests/test_registry_tool_delete_guards.py`:
- Around line 179-214: Replace the predicate-only checks in
TestRegistryToolInUseRefusal with request-level tests that dispatch the
production DELETE route through PromptStudioRegistryView.destroy. Cover owner
deletion success, non-owner deletion denial, and deletion of a workflow-attached
tool returning HTTP 409, asserting response status and relevant side effects so
query, exception, route binding, and destroy permission wiring are exercised.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c35fe762-0e75-4b1c-846e-37a16182b64e

📥 Commits

Reviewing files that changed from the base of the PR and between f41ee0e and 9c9d549.

📒 Files selected for processing (1)

• backend/prompt_studio/prompt_studio_registry_v2/tests/test_registry_tool_delete_guards.py

[chandrasekharan-zipstack]

The registry-delete half holds up on review — I checked the org-isolation argument specifically and it is correct. Not django-tenants (settings/base.py:336 has it commented out, TENANT_APPS = []); it's shared-schema with an organization FK. The viewset inherits OrganizationFilterBackend from DEFAULT_FILTER_BACKENDS (base.py:625-629), DRF's get_object() does call filter_queryset, and that backend fails closed to queryset.none() with no org context. Belt-and-braces, PromptStudioRegistry.objects is itself org-scoped via DefaultOrganizationManagerMixin. So .objects.all() on the detail route does not leak, and the self.kwargs["pk"] vs self.detail reasoning is right.

IsRegistryToolOwner also genuinely mirrors IsParentToolOwner — CustomTool has memberships (prompt_studio_core_v2/models.py:195, backfilled by migration 0009), and the obj.custom_tool or obj fallback resolves to created_by for unlinked rows.

The API-key half has an authorization gap — inline.

[sonarqubecloud]

Quality Gate passed

Issues
19 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[github-actions]

Unstract test results

Per-group results

Status	Group	Tier	Passed	Failed	Errors	Skipped	Duration (s)
✅	e2e-api-deployment	e2e	3	0	0	0	20.8
✅	e2e-coowners	e2e	1	0	0	0	1.4
✅	e2e-etl	e2e	1	0	0	0	8.3
✅	e2e-login	e2e	2	0	0	0	1.3
✅	e2e-prompt-studio	e2e	1	0	0	0	4.4
✅	e2e-smoke	e2e	2	0	0	0	1.1
✅	e2e-workflow	e2e	1	0	0	0	18.1
✅	integration-backend	integration	205	0	0	26	40.9
✅	integration-connectors	integration	1	0	0	7	7.7
✅	integration-workers	integration	140	0	0	1	47.3
✅	unit-backend	unit	340	0	0	1	38.4
✅	unit-connectors	unit	63	0	0	0	10.4
✅	unit-core	unit	33	0	0	0	1.4
✅	unit-platform-service	unit	15	0	0	0	2.7
✅	unit-rig	unit	109	0	0	0	5.4
✅	unit-sdk1	unit	480	0	0	0	24.3
✅	unit-workers	unit	1312	0	0	0	101.1
	TOTAL		2709	0	0	35	335.1

Critical paths

⚠️ Critical paths not yet covered

• workflow-execution-fan-out — Multi-file workflow execution fans out to file-processing workers and rejoins. (declared coverage: no groups declared)

✅ Covered critical paths

• auth-login — covered by e2e-login
• adapter-register-llm — covered by integration-backend
• workflow-author — covered by integration-backend
• co-owner-manage — covered by integration-backend, e2e-coowners
• workflow-create-execute — covered by e2e-workflow
• api-deployment-provision — covered by integration-backend
• api-deployment-auth — covered by integration-backend
• api-deployment-run — covered by e2e-api-deployment
• prompt-studio-author — covered by integration-backend
• prompt-studio-fetch-response — covered by e2e-prompt-studio
• connector-register-test — covered by integration-backend
• pipeline-etl-execute — covered by e2e-etl
• usage-aggregate-read — covered by integration-backend
• usage-token-tracking — covered by e2e-api-deployment
• callback-result-delivery — covered by e2e-api-deployment