rossum-agent 1.0.0rc0__py3-none-any.whl

rossum_agent/skills/rossum-deployment.md

@@ -0,0 +1,102 @@
+# Rossum Deployment Skill
+
+**Goal**: Deploy configuration changes safely via sandbox with before/after diff.
+
+## Credential Identification
+
+**Identify tokens BEFORE any deployment operation.** User may provide tokens with unclear naming.
+
+| Identification Method | How to Apply |
+|-----------------------|--------------|
+| Check organization ID | Use `get_organization(org_id)` with each token to see org name - sandbox orgs typically contain "sandbox", "test", or "dev" |
+| Check org URL pattern | Sandbox orgs often use different base URLs or have distinct naming conventions |
+| Ask user explicitly | If tokens are ambiguous, ask: "Which token is for production and which for sandbox?" |
+| Default env token = production | Token in `ROSSUM_API_TOKEN` env var is production (your main connection) |
+
+**Constraint**: Never assume which token is prod/sandbox. Verify or ask.
+
+## Credential Rules
+
+**Every tool call touches exactly one environment. Know which one before calling.**
+
+| Call Pattern | Environment Affected |
+|--------------|---------------------|
+| `get_schema(...)`, `update_schema(...)`, any direct MCP tool | **PRODUCTION** |
+| `call_on_connection("sandbox", "update_schema", ...)` | Sandbox |
+| `deploy_copy_workspace(..., target_token=SANDBOX_TOKEN)` | Copies FROM production TO sandbox |
+| `deploy_pull(..., token=SANDBOX_TOKEN)` | Pulls FROM sandbox |
+| `deploy_pull(..., token=PROD_TOKEN)` or no token | Pulls FROM production |
+| `deploy_to_org(..., target_token=SANDBOX_TOKEN)` | Deploys TO sandbox |
+| `deploy_to_org(..., target_token=PROD_TOKEN)` | **Deploys TO PRODUCTION** |
+
+**Sandbox modifications**: Always use `call_on_connection("sandbox", tool_name, args)`. Direct MCP calls modify production.
+
+## Workflow
+
+Execute steps 1-5 autonomously. **Only pause at step 5 for user approval.**
+
+| Step | Tool | Token/Connection |
+|------|------|------------------|
+| 1. Copy to sandbox | `deploy_copy_workspace` | `target_token=SANDBOX_TOKEN` |
+| 2. Pull BEFORE | `deploy_pull` → `./before` | `token=SANDBOX_TOKEN` |
+| 3. Modify sandbox | `call_on_connection("sandbox", ...)` | Via spawned connection |
+| 4. Pull AFTER | `deploy_pull` → `./after` | `token=SANDBOX_TOKEN` |
+| 5. Compare + show diff | `deploy_compare_workspaces` → display to user | **Wait for approval** |
+| 6. Deploy to prod | `deploy_to_org` | `target_token=PROD_TOKEN` |
+
+Step 3: Never use direct MCP calls (`update_schema`, `update_queue`, etc.) - those modify production.
+
+## Key Constraints
+
+| Constraint | Rule |
+|------------|------|
+| Pull before compare | **Never call `deploy_compare_workspaces` without first pulling both directories via `deploy_pull`**. Comparison requires local JSON files - there is nothing to compare without pulling first. |
+| Pull BEFORE immediately | After `deploy_copy_workspace`, immediately run `deploy_pull` to `./before`. This captures the baseline. |
+| Pull AFTER last | After all sandbox modifications, run `deploy_pull` to `./after`. Then compare. |
+| Spawned connections don't persist | Re-spawn `spawn_mcp_connection` each conversation turn. |
+| Never deploy without approval | Always show diff and wait for explicit user approval. |
+| IDs differ between environments | Sandbox copies have NEW IDs. Use `deploy_copy_workspace` return value or `call_on_connection("sandbox", "list_queues", ...)` to get sandbox IDs. Production IDs will 404 in sandbox. |
+| Identify credentials first | Before deployment, verify which token is production vs sandbox (see Credential Identification). |
+
+## Sandbox Connection Setup
+
+```python
+spawn_mcp_connection(connection_id="sandbox", api_token="<SANDBOX_TOKEN>", api_base_url="https://api.elis.rossum.ai/v1")
+```
+
+## Common Sandbox Operations
+
+| Operation | Correct (Sandbox) | Wrong (Production) |
+|-----------|-------------------|-------------------|
+| Update schema | `call_on_connection("sandbox", "update_schema", '{"schema_id": 123, ...}')` | `update_schema(schema_id=123, ...)` |
+| Create schema | `call_on_connection("sandbox", "create_schema", '{"name": "...", "content": [...]}')` | `create_schema(name="...", ...)` |
+| Update queue | `call_on_connection("sandbox", "update_queue", '{"queue_id": 456, ...}')` | `update_queue(queue_id=456, ...)` |
+| Update hook | `call_on_connection("sandbox", "update_hook", '{"hook_id": 789, ...}')` | `update_hook(hook_id=789, ...)` |
+
+All read operations (get_schema, get_queue, list_hooks) on sandbox also require `call_on_connection`.
+
+## Tools
+
+| Tool | Purpose |
+|------|---------|
+| `deploy_copy_workspace` | Copy workspace to target org |
+| `deploy_pull` | Pull workspace config (schemas, queues, hooks) to local directory as JSON files |
+| `deploy_compare_workspaces` | Diff two local workspace directories, returns structured changes |
+| `deploy_to_org` | Deploy to target org (`dry_run=True` first) |
+
+## Before/After Diff is Mandatory
+
+**`deploy_compare_workspaces` compares local JSON files, not remote APIs.** Without pulling, there are no files to compare.
+
+| Step | Command | Output Directory |
+|------|---------|------------------|
+| 1. Pull baseline | `deploy_pull(org_id=..., workspace_path="./before", token=SANDBOX_TOKEN)` | `./before/` |
+| 2. Make modifications | `call_on_connection("sandbox", ...)` | - |
+| 3. Pull modified state | `deploy_pull(org_id=..., workspace_path="./after", token=SANDBOX_TOKEN)` | `./after/` |
+| 4. Compare | `deploy_compare_workspaces(source_workspace_path="./before", target_workspace_path="./after")` | Diff output |
+
+**Constraint**: Steps 1 and 3 are prerequisites for step 4. Skipping pull = empty comparison = deployment failure.
+
+**Always show the diff output to the user before deployment.** This is the user's only chance to review changes before they go to production.
+
+Do NOT create markdown files for diffs unless user requests.

rossum_agent/skills/schema-patching.md

@@ -0,0 +1,61 @@
+# Schema Patching Skill
+
+**Goal**: Add, update, or remove individual schema fields.
+
+## Tool
+
+```
+patch_schema_with_subagent(schema_id="12345", changes='[{"action": "add", "id": "invoice_number", "parent_section": "header_section", "type": "string", "label": "Invoice Number"}]')
+```
+
+Sub-agent handles fetching, applying, and verifying changes.
+
+## Changes Format
+
+Each change object in the `changes` array:
+
+| Field | Required | Description |
+|-------|----------|-------------|
+| `action` | No | "add" (default) or "remove" |
+| `id` | Yes | Field ID (schema_id) |
+| `parent_section` | For add | Section ID to add field to |
+| `type` | For add | string, number, date, enum |
+| `label` | No | Defaults to id |
+
+## Field Types
+
+| Type | Extra Fields |
+|------|--------------|
+| `string` | `default_value`, `constraints` |
+| `number` | `default_value` |
+| `date` | `format` |
+| `enum` | `options: [{"value": "v1", "label": "Label 1"}]` |
+
+Not supported: multiline fields. Ignore multiline requests - use regular `string` type instead.
+
+## UI Configuration
+
+Optional `ui_configuration` object controls field behavior in the UI. Only set properties when explicitly requested - do not add ui_configuration if the user hasn't specified type or edit behavior.
+
+| Property | Valid Values | Default |
+|----------|--------------|---------|
+| `type` | `captured`, `data`, `manual`, `formula`, `reasoning`, `null` | `null` |
+| `edit` | `enabled`, `enabled_without_warning`, `disabled` | `enabled` |
+
+Type meanings:
+- `captured` - Value extracted by AI/OCR from document
+- `data` - Value filled by extensions (no bounding box)
+- `manual` - User-entered value (no bounding box)
+- `formula` - Computed from formula definition
+- `reasoning` - Updated per prompt and context
+- `null` - Unset, behaves like captured
+
+Common patterns:
+- Formula field: `{"type": "formula", "edit": "disabled"}`
+- Read-only captured field: `{"type": "captured", "edit": "disabled"}`
+- Extension-filled field: `{"type": "data"}`
+
+## Cross-Reference
+
+- Schema customization during queue creation: load `organization-setup` skill
+- Sandbox testing before production: load `rossum-deployment` skill

rossum_agent/skills/schema-pruning.md

@@ -0,0 +1,23 @@
+# Schema Pruning Skill
+
+**Goal**: Remove unwanted fields from schema in one call.
+
+## Tool
+
+```
+prune_schema_fields(schema_id=12345, fields_to_keep=["invoice_number", "invoice_date", "total_amount"])
+```
+
+Returns `{removed_fields: [...], remaining_fields: [...]}`.
+
+## Behavior
+
+- Specify leaf field IDs only
+- Parent containers (sections, multivalues, tuples) preserved automatically
+- Sections with no remaining children are removed automatically (API rejects empty sections)
+- Alternative: use `fields_to_remove` to remove specific fields instead
+
+## Cross-Reference
+
+- Adding fields after pruning: load `schema-patching` skill
+- Queue creation: load `organization-setup` skill

rossum_agent/skills/ui-settings.md

@@ -0,0 +1,45 @@
+# UI Settings Skill
+
+**Goal**: Update queue UI settings (`settings.annotation_list_table.columns`) without corrupting structure.
+
+## Workflow
+
+1. Fetch current settings via `get_queue`
+2. Modify only `columns` array, preserve all other keys
+3. Patch via `update_queue(queue_id, queue_data={"settings": settings})`
+
+## Column Types
+
+| Type | Required Fields | Optional |
+|------|-----------------|----------|
+| `meta` | `column_type`, `meta_name`, `visible` | `width` (default: 170.0) |
+| `schema` | `column_type`, `schema_id`, `data_type`, `visible` | `width` |
+
+### Meta `meta_name` Values
+
+`status`, `original_file_name`, `labels`, `assignees`, `queue`, `details`, `created_at`, `modified_at`, `confirmed_at`, `exported_at`, `rejected_at`, `deleted_at`, `assigned_at`, `modifier`, `confirmed_by`, `exported_by`, `export_failed_at`, `rejected_by`, `deleted_by`
+
+### Schema `data_type` Values
+
+`string`, `number`, `date`
+
+## Constraints
+
+| Rule | Rationale |
+|------|-----------|
+| `width` must be float | `170.0` not `170` |
+| Validate `schema_id` exists | Check queue schema before adding |
+| Preserve column order | Unless explicitly reordering |
+
+## Default Columns
+
+These columns are always present in the UI by default. To hide them, set `visible: false` explicitly—omitting them from the list does NOT hide them:
+
+`deleted_at`, `labels`, `created_at`, `created_by`, `modified_at`, `modified_by`, `deleted_by`, `rejected_at`, `rejected_by`, `confirmed_at`, `confirmed_by`, `exported_at`, `export_failed_at`, `exported_by`
+
+## Column List Behavior
+
+| User Request | Action |
+|--------------|--------|
+| "Add column X" | Keep existing columns, append new |
+| Provides full list | Include listed columns with `visible: true`, set unlisted default columns to `visible: false` |

rossum_agent/streamlit_app/__init__.py

@@ -0,0 +1 @@
+"""Streamlit test-bed application for the Rossum Agent."""