Skip to content
Back to skills

Synapse

ASecurity

Codex-led multi-model workflow (Codex + Claude + Gemini). Use when the user asks to run `synapse <cmd> ...` to generate draft diffs/audits and persist artifacts to `./.synapse/**`.

  • 40 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added February 8, 2026
developmentpythongoshellnodegitfrontendbackendfullstack

Works with

  • cli
  • mcp

Security analysis

A100/100

Pro scans all 20 files and shows the line behind each finding

Scanned February 12, 2026

npx -y skills add snakeying/Synapse --skill synapse --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Synapse?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Synapse
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/snakeying-synapse/badge)](https://www.skillsdirectory.com/skills/snakeying-synapse)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: synapse
description: Codex-led multi-model workflow (Codex + Claude + Gemini). Use when the user asks to run `synapse <cmd> ...` to generate draft diffs/audits and persist artifacts to `./.synapse/**`.
---

# Synapse (Codex Skill)

Synapse is a **project-local workflow runner** driven by **Codex (main controller)**.

## Models (roles)

- **Codex (main controller)**: decides routing, writes prompts, applies final code, runs verification, and delivers
- **Claude**: planning/architecture/risks + backend draft diffs + general code audit
- **Gemini**: frontend/UI draft diffs + UI/UX + accessibility audit (only when `task_type` includes frontend)

Artifacts are written under `./.synapse/**` in the *target project root*.

## Hard rules (must follow)

- Only handle requests that start with `synapse <cmd> ...` (no slash commands).
- Before running anything: open `references/<cmd>.md` and summarize (purpose, usage, models, writes/side effects, and any safety/confirmation requirements).
- If any **script-level** CLI args are unclear (`init/pack/plan/run/verify/ui`), run `synapse.py <cmd> --help` (via `uv`) and follow the output.
- Run Synapse scripts via `uv`:

```powershell
uv run --no-project python <SKILL_DIR>/scripts/synapse.py --project-dir <PROJECT_DIR> <cmd> ...
```

- Treat any external diff as a **draft**. Do **not** apply verbatim; Codex rewrites to production quality.
- Prompts are **not hardcoded in scripts**. Codex must generate prompts and pass them via `synapse run --prompt-file ...`.
- `workflow`/`feat` are end-to-end (Codex-orchestrated) meta commands. Treat `references/workflow.md` as the source of truth for the exact steps and Gate requirements (including `gate_prep` + single-round Gate).
- `task_type` rules:
  - If the user explicitly sets `--task-type`, respect it.
  - If omitted: default to `fullstack` (info-complete; higher cost) and still present `frontend/backend/fullstack` options + Codex recommendation at the Gate.
- Git safety:
  - Prefer a clean working tree. If it's dirty, surface it at the Gate and ask how to proceed (stash/commit/abort).
  - If the project is not a git repo, propose `git init` at the Gate (review quality is best with git).

## Global flags (common)

- `--project-dir <dir>`: directory inside the target project (default: `.`)
- `--resume-gemini <SESSION_ID>` / `--resume-claude <SESSION_ID>`
- `--resume <SESSION_ID>`: alias for `--resume-gemini` (back-compat)
  - If both `--resume` and `--resume-gemini` are provided with different values, Synapse errors to avoid silent confusion.

## Command map

User-level commands (typed in Codex chat):

- `workflow`, `feat`

Script-level subcommands (Codex runs via `uv`):

- `init`, `pack`, `plan`, `run`, `verify`, `ui`

## Script guarantees

- Project root detection: `git rev-parse --show-toplevel` (fallback: current directory).
- Artifacts always go under `./.synapse/**` (plans/context/logs/patches/state/index).
- Writes are guarded by `assets/defaults.json:safety.allowed_write_roots` (hard fail if a write would escape these roots).
- `init` only writes: `AGENTS.md`, `.gitignore`, `./.synapse/**` (idempotent).
- `verify` may create **project-local toolchain artifacts** (lockfiles, `.venv/`, `node_modules/`, build outputs) as a side effect.
- External model calls use local `claude` + `gemini` CLIs in stream-json mode (`session_id` captured; resume supported). Synapse never enables “auto-approve” modes (Gemini runs with `--approval-mode default`; Claude runs with `--permission-mode plan --tools "" --strict-mcp-config`).
- `ui` starts a local read-only web viewer for `.synapse/**` (prompts/outputs/patches/logs/state).

Files in this skill

  • SKILL.md3.6 KB
  • assets/defaults.json768 B
  • assets/ui/index.html21.7 KB
  • references/feat.md213 B
  • references/init.md437 B
  • references/pack.md472 B
  • references/plan.md1.2 KB
  • references/run.md1.3 KB
  • references/ui.md1 KB
  • references/verify.md605 B
  • references/workflow.md2.5 KB
  • scripts/_synapse/__init__.py61 B
  • scripts/_synapse/agents_md.py3.5 KB
  • scripts/_synapse/cmd_init.py1 KB
  • scripts/_synapse/cmd_pack.py2.5 KB
  • scripts/_synapse/cmd_plan.py5.7 KB
  • scripts/_synapse/cmd_run.py5.7 KB
  • scripts/_synapse/cmd_ui.py5.5 KB
  • scripts/_synapse/cmd_verify.py5.8 KB
  • scripts/_synapse/common.py9 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.