One AI-insider selection rule for the page and the MCP server

[mrwadams]

Automated by agent-loop — the agent worked this issue on a fresh clone of mrwadams/attackgen inside a disposable sandbox; changes were gated outside the agent.

Closes #112

How to test

git fetch origin && git checkout agent/issue-112
python3 -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt
streamlit run "00_👋_Welcome.py"

Then open the printed local URL.

Confirm each acceptance criterion:

• With categories selected and no STRIDE threats, the MCP AI-insider tools and the page produce the same prompt, including the "Specific STRIDE threats in scope" block.
• Omitting agent capabilities yields the full capability set; passing an empty list yields the deliberate opt-out line. Both hold for the page and for both MCP AI-insider tools.
• Category-to-STRIDE derivation and the capability default each exist in exactly one place, behind build_ai_insider_messages.
• The page's derivation block and the MCP server's capability default are gone.
• The test that asserts a literal default string appears in the page's source text is deleted, replaced by a test comparing the two entry points' prompts for the same selection.
• Prompts are byte-identical to today for an existing page generation and for an MCP call that passes explicit STRIDE threats.
• pytest is green, including the pinned prompt-template tests.

Gates (non-authoritative)

• ✅ files_non_empty — 4 non-empty file(s) changed
• ✅ containment — diff stays within the allowed lane
• ✅ verify — pytest: 333 passed, no failures (browser/e2e excluded)

Draft. Browser/e2e tests were NOT run in-gate — run them and review before merging. Nothing here is auto-merged.