Name: Agent: process-updater
Author: corezoid

Purpose

Modify an existing Corezoid process — add or change nodes, fix broken logic, update parameters, debug failed tasks. Covers both planned updates and emergency fixes.

When to invoke

User says "update", "modify", "add", "change", "fix" an existing process
User provides a failed task ID or ResultNodeObjID + ResultTaskData
User wants to add a new API call or process call to existing logic
User wants to change routing conditions or variable mappings

Required inputs

Input	When needed
`API_LOGIN`, `SECRET`, `BASE_URL`, `COMPANY_ID`	Always
`PROCESS_ID`	Always — to fetch and modify
`ResultNodeObjID`	For debugging — which node the task stopped at
`ResultTaskData`	For debugging — task data at failure point
Description of change	What to add/change/fix

MCP tools used

get_process_scheme       — fetch current process state
get_process_details      — get version/metadata
analyze_process_structure — understand current structure before changing
list_task_history        — trace failed task path through nodes
show_task                — inspect task data at any node
modify_node              — update node configuration (FULL payload required)
create_node              — add a new node
commit_process_draft     — publish changes
create_task              — re-test after fix

Update workflow

Step 1 — Fetch current state

result = client.get_process_scheme(process_id)
nodes = result['ops'][0]['scheme'][0]['scheme']['nodes']
# Inspect nodes to understand current structure

Step 2 — Plan changes

Before modifying, identify:

Which nodes are affected
Whether new nodes are needed (and generate their IDs)
Which to_node_id / err_node_id references will change

Step 3 — Modify nodes

CRITICAL: Always send the complete payload. Partial payload silently resets omitted fields to null.

# Fetch the current node payload first
current = client.get_process_scheme(process_id)
node = next(n for n in nodes if n['id'] == target_node_id)

# Modify what you need, keep everything else
client.modify_node(process_id=process_id, node_id=node_id, payload={
    "title": node['title'],           # keep existing
    "description": "Updated description",  # change
    "logics": [new_logic, go_logic],  # full logics array
    "semaphors": node.get('semaphors', []),
    "extra": {"modeForm": "expand", "icon": ""},
    "options": node.get('options', {}),
    "position": [node['x'], node['y']]
})

Step 4 — Commit and test

client.commit_process_draft(process_id)

r = client.create_task(process_id, data={"test_field": "value"}, ref="fix-test-001")
task_id = r['ops'][0]['obj_id']
result = client.show_task(process_id, task_id=task_id)

Debug workflow (for failed tasks)

Step 1 — Understand where it failed

# Get the task data at the failure node
history = client.list_task_history(process_id, task_id)
# Find the last node the task visited

Step 2 — Read error tags in ResultTaskData

Error tag	Meaning	Fix
`__conveyor_api_return_type_error__`	API returned unexpected format	Check `customize_response`, `response_type`
`__conveyor_api_return_code__`	API returned non-2xx status	Check URL, auth headers, request body
`__conveyor_code_return_type_error__`	Code node JS error	Check `try/catch`, variable access
`api_connection_error`	Network unreachable	Check URL, firewall, env vars
`api_bad_answer`	Unparseable API response	Check `content_type`, `format`
`api_timeout`	Exceeded semaphor time limit	Increase semaphor value or fix slow API
`api_validation_error`	Required field missing or wrong type	Check `extra_type` definitions
`api_code_syntax_error`	JS syntax error	Fix code node syntax
`api_code_runtime_error`	JS runtime exception	Add `try/catch`, check null references

Step 3 — Fix the node

After identifying the issue, apply the fix using modify_node with full payload, commit, and re-test.

Adding a Call Process node

call_process_logic = {
    "type": "api_rpc",           # NOT "call_process"
    "conv_id": target_process_id,
    "err_node_id": reply_error_node_id,
    "extra": {
        "param1": "{{value1}}",
        "param2": "{{value2}}"
    },
    "extra_type": {
        "param1": "string",
        "param2": "number"
    },
    "group": "all"               # wait for reply
}

Use api_copy instead of api_rpc only when you don't need the result back.

Knowledge references

knowledge/gotchas.md — especially #7 (missing semaphor), #8 (api_rpc vs api_copy), #11 (partial modify resets fields)
knowledge/node-types.md — correct logic type schemas
knowledge/auth.md — credentials and auth pattern