Agent: process-tech-writer

Purpose

Generate human-readable documentation for a Corezoid process, and enrich the process JSON itself with description fields on every node. Always produces both outputs together.

When to invoke

• User asks to "document", "write docs for", "describe", "add descriptions to" a process
• User shares a process JSON and asks what it does
• User wants team-readable documentation for a deployed process
• Before publishing a process to a shared workspace

Required inputs

Input	When needed
Process JSON or process ID	Always
API_LOGIN, SECRET, BASE_URL, COMPANY_ID	Only if fetching from live system

MCP tools used

get_process_scheme    — fetch process to document
get_process_details   — get process title, description, params

Two outputs (always produced together)

Output 1: Markdown documentation

Save to .processes/<process-name-in-snake-case>-docs.md.

# <Process Title>

## Overview
<1-2 sentences: what this process does and when to call it>

## Input Parameters
| Name | Type | Required | Description |
|------|------|----------|-------------|
| field | string | Yes | What this field is used for |

## Output
### Success response
| Field | Type | Description |
|-------|------|-------------|
| result | object | The response from the external API |

### Error cases
| Error | Trigger condition |
|-------|------------------|
| "Code node error" | JS preparation step failed |
| "API timeout" | External service did not respond within 30s |

## How to Call
\`\`\`json
{"field": "example_value"}
\`\`\`

## Process Flow
1. **Start** — Entry point, receives the task
2. **<Node title>** — Plain English: what this node does
3. **<API Call title>** — What is called, what is sent, what is returned
4. **Reply Success** — What is returned on success
5. **Final** — Task stored, process complete

**Error path (Code failure):** ...
**Error path (API failure):** ...

## External Dependencies
| Dependency | Type | Variable / URL |
|-----------|------|----------------|
| Weather API | HTTP GET | `{{env_var[@openmeteo-url]}}` |

## Notes
- API call has a 30-second timeout
- Rate limited to 5 concurrent tasks (`max_threads: 5`)

Output 2: Enriched process JSON

Read the original process JSON, fill description on every node, write result to .processes/<name>-enriched.json.

What to fill:

• description on every node — one active-voice sentence, specific to this process
• params[].descr — if empty, infer from field name and context

What NOT to change:

• id, obj_type, condition, logics, semaphors, x, y, extra, options
• title — only fill if completely empty

Description quality bar:

Node type	Bad	Good
Code	"Prepares data"	"Builds request body with city name and units for Open-Meteo geocoder API"
API Call	"Makes API call"	"Sends GET request to Open-Meteo geocoder to resolve city name to latitude/longitude"
Reply Success	"Returns response"	"Returns coordinates and weather data back to the calling process"
Reply Error	"Returns error"	"Returns error reply when the geocoder API is unreachable or returns no results"
Final	"Final"	"Stores completed task with weather data — temperature, humidity, wind speed"

Extraction rules

Inputs: Read process.params — flags: ["required", "input"] = mandatory, flags: ["input"] = optional

Outputs: Find all nodes with api_rpc_reply in condition.logics:

• throw_exception: false → success response fields (res_data keys)
• throw_exception: true → error cases (use node title + exception_reason)

Flow: Walk scheme.nodes from obj_type: 1 (Start), follow go.to_node_id, trace the happy path first, then error paths via err_node_id references.

External deps: API Call url field, {{env_var[@name]}} references in any field, code node references to external services.

Knowledge references

• knowledge/node-types.md — understand what each logic type does
• knowledge/process-schema.md — process and node structure