v3.2.3 — Python library API layer

What changed

Root cause fixed: import synapse_forge resolved to the package directory (synapse_forge/__init__.py) rather than the standalone synapse_forge.py file. The __init__.py was a 9-line stub that added the parent to sys.path and nothing else — leaving plan, compose, dispatch, run, call_llm, and all registry helpers absent from the imported module.

Changes

synapse_forge/__init__.py (rewritten — +281 LOC)

• Loads synapse_forge.py as _forge_core via importlib.util
• Re-exports all public names into the package namespace
• Defines thin wrappers for dispatch, call_llm_with_fallback, and run in the package's own __dict__ so patch("synapse_forge.call_llm") intercepts correctly in tests (mock patching works by replacing the name in the target module's dict; wrappers defined here resolve call_llm through this module's globals at call time)
• Exposes detect_providers from synapse_forge/detect.py

synapse_forge.py (+3 LOC changed)

• plan_id now uses time.time_ns() instead of int(time.time() * 1000) — guarantees uniqueness across rapid same-millisecond calls
• Version string bumped to 3.2.3 in docstring, main() parser, and status output

Test results

Before	After
14/51 passing	51/51 passing

Backwards compatibility

• CLI invocation (python synapse_forge.py run "...", synapse-cor run "...") unchanged
• Adapter imports (from synapse_forge.adapters.gemini import Adapter) unchanged
• install.sh / bootstrap.sh flow unchanged

v3.2.2 — Full adapter prerequisite coverage

What's new

All 7 adapters now return structured prerequisite-check dicts so the smart installer can give precise per-provider guidance. Plus several artifacts that subagents had built but not committed in v3.2.1 are now in.

Added

• check_prerequisites() on codex, gemini, generic-openai, openswarm, synapse-swarm adapters (claude-code + gemini-cli-free already had it in v3.2.1)
• sentinel/ package — portable error-handling/self-healing daemon (3 detectors: silent-failure, model-stall, hallucination-drift)
• synapse_swarm.py — portable session daemon that's API-compatible with OpenSwarm but routes through the bundle's cascade (replaces OpenSwarm v2.0.8 which has a known modelID-routing bug)
• adapters/synapse-swarm/ — drop-in replacement for the openswarm adapter, with graceful fallback when daemon is down

Notes

• The openswarm adapter now warns about the v2.0.8 modelID bug and recommends synapse-swarm
• Coverage: 7/7 adapters with check_prerequisites() ✓

🤖 Generated with Claude Code

v3.2.1 — Zero-context user audit

What's new

The "fresh user with only Claude Code installed" experience is now first-class. No assumed context, no FUTRON-specific prerequisites, full guidance for naive users.

Added

• BOOTSTRAP_GUIDE.md — per-OS Python+Node install walkthroughs, 4 LLM-provider paths (Gemini-free, Ollama, Groq, key-required)
• bootstrap.sh — optional one-shot setup (auto-detects OS, installs Node via nvm, installs gemini-cli)
• docs/glossary.md — plain-language definitions: Synapse_COR, Forge, MCC, CoR, cascade, deterministic-vs-probabilistic
• docs/troubleshooting.md — common errors → fixes
• docs/use-cases.md — 6 detailed examples with bash commands and sample output
• CONTRIBUTING.md + CODE_OF_CONDUCT.md — standard OSS files
• README 30-second TL;DR at top, fixed clone URL bug

Changed

• Adapter check_prerequisites() method added to gemini-cli-free + claude-code adapters (more in v3.2.2)

🤖 Generated with Claude Code

v3.2.0 — Zero-Config Smart Installer + Gemini CLI Free-Mode

What's New in v3.2.0

Zero-config first run

The bundle now works out of the box. Run `bash install.sh` and it auto-detects available LLM providers — no `SYNAPSE_FORGE_URL` required.

Provider auto-detection (`synapse_forge/detect.py`)

Scans the system in cheapest-first order:

1. Local Ollama (:11434)
2. Local LM Studio / vLLM (:1234, :8000, :5000)
3. Gemini CLI OAuth (`~/.gemini/google_accounts.json`)
4. Gemini CLI `--free` flag — no login, no API key (unofficial endpoint)
5. `GEMINI_API_KEY` / `OPENAI_API_KEY` / `GROQ_API_KEY` / `OPENROUTER_API_KEY`
6. Codex OAuth (`~/.codex/auth.json`)
7. Generic `*_API_KEY` sweep

Gemini CLI free-mode adapter (`adapters/gemini-cli-free/`)

Subprocess adapter that tries `gemini-cli --free` first (no auth), falls back to keyless OAuth, returns a clear install-guide message if both fail.

Smart installer rewrite (`install.sh`)

• Runs detection, writes `~/.synapse-forge/auth-config.json`
• If no providers found: prints onboarding steps for each free path
• Optional interactive key prompt (skip with `--non-interactive` for CI)

call_llm_with_fallback cascade (synapse_forge.py)

All dispatch calls now auto-cascade: explicit URL → Ollama → Gemini free → OAuth → paid API keys. `SYNAPSE_FORGE_URL` is now optional.

New CLI subcommand

```bash
synapse-cor detect # show all detected providers ranked by cost
synapse-cor status # now includes provider count and best provider
```

Upgrade

```bash
git pull && bash install.sh
```

v3.1.0 — Initial Public Release

What is Synapse_COR?

A token-efficient AI orchestration kernel that takes a goal and decomposes it into a PLAN.json separating deterministic CLI steps (which need no LLM) from probabilistic LLM steps (which do). Each step routes to the cheapest capable model. Specialist agents are spawned with scoped personas, MCC cognitive stacks, and behavioral contracts.

v3.1.0 Highlights

• 40 self-contained tests, 100% pass on Python 3.9-3.12
• 5 platform adapters: Claude Code, Codex, Gemini, OpenSwarm, generic-OpenAI
• Full CoR template (18 fields)
• PyPI-ready (pyproject.toml + setup.py)
• GitHub Actions CI matrix
• MIT licensed

Quick Install

```bash
git clone https://github.com/FutronPrime/synapse-cor
cd synapse-cor
bash install.sh
export PATH="$HOME/.synapse-forge/bin:$PATH"
synapse-cor run "audit my project for security issues"
```

Documentation

• Quickstart
• Architecture
• Extending