fix(clone): clone agentic project documents + resolve exported agentic tools

[chandrasekharan-zipstack]

What

Completes org-to-org clone for cloud Agentic ("Agentic Prompt Studio") projects, which previously cloned only the project shell:

1. Documents — each project's uploaded docs are cloned (download raw bytes from source → skip names already on target → upload to the target project). Honours file_strategy="skip" like the files/lookups phases.
2. Verified data — the curated ground-truth rows are cloned, re-pointed to the cloned document by filename. Extracted/comparison data is regenerable and intentionally not cloned.
3. Exported agentic tools — a workflow tool_instance.tool_id now resolves against agentic_studio_registry in addition to prompt_studio_registry.

Why

• Agentic projects keep uploads in a separate store (agentic/documents/), distinct from Prompt Studio prompt-documents. The files phase only iterates the custom_tool remap, so agentic docs were silently dropped — a clone landed the project with zero documents.
• agentic_verified_data is "ground truth manually verified by user" — curated input (the accuracy baseline), not regenerable output. Without it the target org can't measure extraction accuracy without a human re-verifying every doc.
• A workflow tool_instance.tool_id is a registry id. Exported agentic projects register under agentic_studio_registry, but ToolInstancePhase resolved only via prompt_studio_registry, so an agentic-tool instance found no remap and was skipped (no registry remap for tool_id …). The dependent workflow (e.g. "Agentic tool API") then landed with no tool wired.

All three reproduced on a staging org→org run.

How

• client.py: list_agentic_documents, download_agentic_document (raw binary, like download_lookup_file), upload_agentic_document (multipart to agentic/projects/{id}/documents/upload/ — the real upload route; the documents viewset upload action is a backend stub); list_agentic_verified_data, create_agentic_verified_data.
• agentic_studio.py: _clone_documents / _clone_one_document and _clone_verified_data run after schemas, before registry republish; idempotent by filename; honour max_file_size and file_strategy; dry-run plan counts both.
• tool_instance.py: tool_id resolves via prompt_studio_registry or agentic_studio_registry; corrected the misleading "custom tool unpublished" skip message.

Like Prompt Studio uploads, this clones the file + creates the document row; extraction/summary stays a UI step.

Can this PR break any existing features?

No. New client methods are additive. The doc/verified-data paths are gated to the cloud-only AgenticStudioPhase (probe-skipped on OSS) and add work that previously didn't happen at all. The tool_instance change only adds a fallback resolve when the primary lookup misses, so existing Prompt Studio tool instances are unaffected.

Database Migrations

• None.

Env Config

• None.

Relevant Docs

• N/A

Related Issues or PRs

• Extends the agentic-studio clone support in AgenticStudioPhase.

Dependencies Versions

• None.

Notes on Testing

• tests/clone/ green (192 passing). Added: agentic doc-clone (skip-existing, dry-run count, file_strategy=skip), verified-data clone (filename mapping, skip-existing, skip-when-doc-missing, dry-run count), and tool_instance resolve-via-agentic-registry.
• Pre-existing ruff format/E501 drift in untouched files left alone; changed lines are lint + format clean.

Checklist

I have read and understood the Contribution Guidelines.

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR completes the org-to-org clone for Agentic Prompt Studio projects by adding document upload/download, ground-truth verified-data cloning, and a fallback agentic_studio_registry resolver for workflow tool instances. The three previously flagged issues have all been resolved.

• Document cloning: _clone_documents / _clone_one_document download raw bytes from source and upload to target, correctly short-circuits on file_strategy=\"skip\" and max_file_size.
• Verified-data cloning: _clone_verified_data re-points ground-truth rows to the cloned target document via filename; skips rows whose document is missing, and returns early under file_strategy=\"skip\".
• Tool-instance resolution: ToolInstancePhase now falls back to agentic_studio_registry when the prompt_studio_registry lookup misses.

Confidence Score: 5/5

Safe to merge

All three previously flagged issues are resolved, new methods are additive, and the agentic-registry fallback is inside the existing lock.

No files require special attention.

Important Files Changed

Filename	Overview
src/unstract/clone/client.py	Adds five new client methods following the established request/error-handling pattern.
src/unstract/clone/phases/agentic_studio.py	Adds document and verified-data clone methods with correct file_strategy/dry-run guards.
src/unstract/clone/phases/tool_instance.py	Adds agentic_studio_registry fallback inside the existing lock — thread-safe and correct.
tests/clone/test_agentic_studio_phase.py	Eight new test cases covering all critical edge cases.
tests/clone/test_tool_instance_phase.py	Adds test verifying the agentic-registry fallback resolve path.

_{Reviews (4): Last reviewed commit: "fix(clone): honour skip strategy in _clo..." | Re-trigger Greptile}