Add Formal AI dispatch for every agentic CLI

[konard]

Fixes #2059.

Problem

solve <issue> --tool agent --model formal-ai --only-prepare-command was rejected during model validation. Formal AI was not registered for any Hive Mind tool, and there was no supported persistent Formal AI service for host or Docker-isolated runs.

Solution

• register formal-ai for Agent, Claude, OpenCode, Codex, Qwen, and Gemini
• route the selected tool through formal-ai with, using formalai/formal-ai where Agent/OpenCode require a provider-qualified model
• preserve every tool's native execution path and existing default model
• make --only-prepare-command assemble the real wrapped executor command without invoking a CLI or mutating OpenCode configuration
• validate both the Formal AI wrapper and selected native CLI during connection checks
• install pinned Formal AI 0.305.0 binaries in all Hive Mind runtime images
• add a separate privileged Dockerfile.formal-ai service with standard port 8080 and persistent /home/box/.formal-ai memory
• connect Hive Mind and the sidecar on the explicit link-assistant-formal-ai Docker network/hostname
• forward the persistent endpoint into nested Docker isolation, resolving the outer Compose hostname before crossing Docker-daemon boundaries
• document requirements, architecture, benchmark evidence, and the measurable gate for any future default-model promotion under docs/case-studies/issue-2059/

Reproduction and regression coverage

Before this change, the minimal reproduction was rejected:

solve https://github.com/owner/repo/issues/1 \
  --tool agent \
  --model formal-ai \
  --only-prepare-command

tests/test-issue-2059-formal-ai-dispatch.mjs now covers:

• both Formal AI model identifiers
• wrapper dispatch and public prepared-only execution for all six tools
• native command preservation
• external server flags and wrapper override path
• connection preflight
• nested Docker endpoint propagation and hostname resolution
• Docker image, network, hostname, port, and persistent-memory contracts

Verification

• npm test — all 352 selected test files passed
• focused issue suite — 31/31 passed
• ESLint — passed with zero warnings
• Prettier, MJS syntax, file-line limits, duplication threshold, documentation validation, and language synchronization — passed
• changeset and version checks — passed
• all four Dockerfiles passed docker build --check
• Compose configuration rendered successfully
• Formal AI 0.305.0 built in the pinned Rust 1.96 builder, ran in the Ubuntu 24.04 Hive Mind base, and reported a compatible maximum GLIBC requirement of 2.34
• real wrapper smoke tests accepted all six target CLIs
• live Formal AI server /health returned OK on port 8080

This is not a visual/UI change, so screenshots are not applicable.

[konard]

Working session summary

Completed and ready for review: PR #2108 — Add Formal AI dispatch for every agentic CLI.

Implemented:

• Formal AI dispatch for Agent, Claude, OpenCode, Codex, Qwen, and Gemini.
• Persistent Formal AI sidecar, port 8080, named Docker network, and nested-container routing.
• Runtime-image installation, validation, documentation, changeset, and detailed case study.
• Existing default models remain unchanged pending the documented capability gate.

Verification:

• Focused tests: 31/31 passed.
• Full local suite: all 352 files passed.
• Fresh CI run: all required checks passed, including Docker runtime verification.
• PR is mergeable, CLEAN, current with main, and the worktree is clean.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-5.6 Sol
• Provider: OpenAI
• Public pricing estimate: $94.000600

📊 Context and tokens usage:

• 1.6M / 200K (784%) input tokens, 177.6K / 128K (139%) output tokens

Total: (1.6M + 70.3M cached) input tokens, 177.6K output tokens, $94.000600 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-5.6-sol
• Thinking level: xhigh (~31999 tokens)
• Model: GPT 5.6 Sol (gpt-5.6-sol)

📎 Log file uploaded as Gist (21248KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag