schema_version = "1.4"

[task]
name = "wildawm/f00190-d9-m1"
version = "1.0.0"
description = "Scope 3 ingest launch prep: pipeline analysis, model collection, and team notification"
keywords = ["wildawm", "xai", "hard", "huggingface", "notification_feed", "cli", "mcp", "web"]

[metadata]
difficulty = "hard"
persona = "f00190-taylor-brooks"
day = 9
sites = ["huggingface"]
mcp_domains = ["notification_feed"]
starting_interface = "mcp"
hops = 4
archetype = "data_pipeline_launch"
area = "work"
interface_flow = "mcp: read the Hugging Face notification feed for existing collection references -> cli: parse the RFC and eval logs to build model_analysis.json -> web: create a new Hugging Face collection and add the visually-verified zero-shot classifier -> mcp: post a notification to the feed with the collection name and accuracy summary"
set_val33 = true
set_sft100 = false
set_rl208 = false
set_rl100 = false
set_july_14_100 = false
set_july_30_val = true
set_aug_10 = false
set_sept_08 = false
frozen_clock_date = "2026-04-09"
frozen_clock_time = "10:00:00"
source_task = "tasks_xai/f00190-taylor-brooks/day9/f00190-d9-m1"
world_blob_sha256 = "5604d152c7636ff9d32d47ebb7a32adb05dfb534862f39e9f6a518274545bcd7"

[agent]
timeout_sec = 3600.0

[verifier]
timeout_sec = 900.0

[environment]
build_timeout_sec = 1800.0
cpus = 2
memory_mb = 4096
storage_mb = 20480
workdir = "/workspace"

[environment.env]
FAKETIME = "@2026-04-09 10:00:00"
LD_PRELOAD = "/usr/lib/x86_64-linux-gnu/faketime/libfaketime.so.1"
AWM_MCP_GATEWAY_URL = "http://127.0.0.1:18080"
AWM_BROWSER_MCP_PORT = "8931"
NODE_TLS_REJECT_UNAUTHORIZED = "0"
# Vision model for describe_image; values come from the harbor environment at run time.
WILDAWM_VISION_LLM_ENDPOINT = "${WILDAWM_VISION_LLM_ENDPOINT:-http://host.docker.internal:8000/v1}"
WILDAWM_VISION_LLM_MODEL_NAME = "${WILDAWM_VISION_LLM_MODEL_NAME:-}"
WILDAWM_VISION_LLM_KEY = "${WILDAWM_VISION_LLM_KEY:-}"
# Optional vision tuning.
WILDAWM_VISION_LLM_REASONING_EFFORT = "${WILDAWM_VISION_LLM_REASONING_EFFORT:-none}"
WILDAWM_VISION_LLM_REASONING_CHAT_PARAM = "${WILDAWM_VISION_LLM_REASONING_CHAT_PARAM:-reasoning_effort}"
WILDAWM_DESCRIBE_IMAGE_MAX_TOKENS = "${WILDAWM_DESCRIBE_IMAGE_MAX_TOKENS:-4096}"
WILDAWM_DESCRIBE_IMAGE_TIMEOUT = "${WILDAWM_DESCRIBE_IMAGE_TIMEOUT:-180}"
WILDAWM_DESCRIBE_IMAGE_RETRIES = "${WILDAWM_DESCRIBE_IMAGE_RETRIES:-3}"
WILDAWM_DESCRIBE_IMAGE_LOCK_TIMEOUT = "${WILDAWM_DESCRIBE_IMAGE_LOCK_TIMEOUT:-180}"
# Packaged Playwright browsers.
PLAYWRIGHT_BROWSERS_PATH = "/opt/pw-browsers"

[environment.healthcheck]
command = "bash /opt/wildawm-harbor/healthcheck.sh"
interval_sec = 5.0
timeout_sec = 30.0
start_period_sec = 900.0
start_interval_sec = 5.0
retries = 60

[[environment.mcp_servers]]
name = "wildawm"
transport = "streamable-http"
url = "http://localhost:18080/mcp"

[[environment.mcp_servers]]
name = "email"
transport = "streamable-http"
url = "http://localhost:18080/email/mcp"

[[environment.mcp_servers]]
name = "slack"
transport = "streamable-http"
url = "http://localhost:18080/slack/mcp"

[[environment.mcp_servers]]
name = "calendar"
transport = "streamable-http"
url = "http://localhost:18080/calendar/mcp"

[[environment.mcp_servers]]
name = "canvas_lms"
transport = "streamable-http"
url = "http://localhost:18080/canvas_lms/mcp"

[[environment.mcp_servers]]
name = "direct_messages"
transport = "streamable-http"
url = "http://localhost:18080/direct_messages/mcp"

[[environment.mcp_servers]]
name = "library_system"
transport = "streamable-http"
url = "http://localhost:18080/library_system/mcp"

[[environment.mcp_servers]]
name = "support_tickets"
transport = "streamable-http"
url = "http://localhost:18080/support_tickets/mcp"

[[environment.mcp_servers]]
name = "notification_feed"
transport = "streamable-http"
url = "http://localhost:18080/notification_feed/mcp"

[[environment.mcp_servers]]
name = "browser"
transport = "streamable-http"
url = "http://localhost:8931/mcp"

[[environment.mcp_servers]]
name = "vision"
transport = "stdio"
command = "python3"
args = ["/agent-config/describe_image_mcp.py"]