fix(ops): deploy installs full requirements.txt (incl non-Claude model stack)

The agent-team venv was missing langchain-anthropic/-openai/-google-genai/
-community, so models.py failed to import and the in-process GPT-4.1 review /
Gemini scan / DeepSeek build silently fail-closed to REQUEST_CHANGES (the
non-Claude models never ran on the box). Step 3 now installs the full pinned
requirements.txt into the venv instead of just fastapi/uvicorn. Installed +
verified live on the box: all three model factories construct.
This commit is contained in:
Adam Moussa 2026-06-23 13:37:11 -04:00
parent 75f1dc6f75
commit 478bd7f90b

View file

@ -51,10 +51,15 @@ else
echo "WARN: ${HANDBOOK_LOCAL} not found; context_provider will return '' (fail-safe). Skipping."
fi
say "3. Install new venv deps (fastapi/uvicorn for the WS1 HTTP API)"
# requirements.txt now pins fastapi==0.136.1 / uvicorn==0.46.0. The coordinator
# itself does not need them, but the optional HTTP API (step 6) does.
$SSH 'cd ~/orchestrator/agent-team && . .venv/bin/activate && pip install --upgrade "fastapi==0.136.1" "uvicorn==0.46.0"'
say "3. Install venv deps from the pinned requirements.txt"
# Install the FULL pinned set into the agent-team venv. This includes the
# non-Claude model stack (langchain-anthropic/-openai/-google-genai/-community)
# that the in-process invokers (WS1: GPT-4.1 review, Gemini scan, DeepSeek build)
# import via models.py — WITHOUT these, models.py fails to import and the review
# loop silently fail-closes to REQUEST_CHANGES (the non-Claude models never run).
# Also brings fastapi/uvicorn (WS1 HTTP API). Leaves the venv-only deps that are
# NOT in requirements.txt (claude-agent-sdk, slack_sdk, slack_bolt) untouched.
$SSH 'cd ~/orchestrator/agent-team && . .venv/bin/activate && pip install --upgrade -r ~/orchestrator/requirements.txt'
say "4. Append new secrets to ~/secrev.env if absent (mode 600, never committed)"
# AGENT_TEAM_API_TOKEN: required only if you run the HTTP API / /delegate hook.