Skip to content

USER-GUIDE.md

Latest commit

 

History

History
204 lines (153 loc) · 7.99 KB

USER-GUIDE.md

File metadata and controls

204 lines (153 loc) · 7.99 KB
Raw

Myco User Guide (Windows)

Myco — the mycelial layer for your agents. It connects the AI coding assistants installed on your machine (Claude Code, WorkBuddy, Codex CLI, Cursor, Antigravity…) into one network: share skills, hand off conversations, unify history — while staying strictly read-only toward every agent's own data.

The macOS build has the same features in menu-bar form. Screenshots below are from the Windows build. 中文版：USER-GUIDE.zh-CN.md

Install & launch

  1. Grab either from the latest release (the .NET runtime and a Python engine are bundled in both):
    • Myco-Setup-x.y.z.exe (recommended) — standard wizard: per-user install, Start-Menu shortcut, optional start-at-login, uninstall via Settings → Apps
    • Myco-win-x.y.z.zip (portable) — unzip anywhere, double-click Myco.exe
  2. SmartScreen may warn on first run (unsigned): More info → Run anyway, once.
  3. Myco lives in the system tray (bottom-right, the stacked-tiles icon):
    • Left click — open / dismiss the panel
    • Right click — Open · Re-detect · Quit

The panel dismisses itself when you click elsewhere, like the system calendar flyout.

Requirements: Windows 10/11. On Win11 the panel gets a native acrylic glass backdrop.

Home — your agents at a glance

Home tab

The first screen when the panel opens:

Area Meaning
Neon-lime hero card Total sessions available for archiving / handoff, across every installed agent
Two small cards Detected agent count · distributable skill count
AGENTS list One row per agent: colored initial badge + data directory + status
Bottom cards Shortcuts into Share and Relay

Status badges:

  • 约 N 段 (≈ N sessions) — installed; count comes from counting *.jsonl files, hence approximate
  • 未安装 (not installed) — no data directory for that agent on this machine
  • 结构已变 (layout changed, amber) — the agent is installed but its session storage no longer matches what Myco knows (usually a product update). Myco says so honestly instead of showing a wrong number.

Detection paths are driven by engine/agents.json — adding or fixing an agent is a one-file edit.

Share — write a skill once, every agent can use it

Share tab

Fans a skill (a SKILL.md directory) out into each agent's conventional skill directory; after git commit, every tool on the team can read it.

  1. Pick the source skill — the top card shows the current one (a switch button appears when there are several)
  2. Tick the targets — the list comes from agents.json: five agent directories (.claude/skills, .codex/skills, …) plus two generic ones (.agents/skills — cross-agent, recommended; .cline/skills)
  3. Dry-run is on by default — it only lists the paths that would be written, touching nothing. Turn it off and click again to actually write
  4. Click 分发 skill (Distribute); the result card lists what happened

Don't forget to git commit afterwards — a skill that isn't in Git isn't shared.

Relay — switch tools, keep the conversation

Relay tab

Packages one conversation from product A as paste-ready text so you can continue it in product B. No forged session IDs, no writing into any database — product B continues in its own, legitimately created session.

  1. The list shows real sessions from every agent (title · short id · turns · date); pick one
  2. Choose a packaging mode:
    • 自动 (auto) — decided by conversation length (recommended)
    • 完整 (full) — the whole transcript, most faithful
    • 摘要 (summary) — condensed, for very long chats
    • 近期 (recent) — only the latest exchanges
  3. Click 生成接力包 (Build handoff) — the package is written to the output directory (handoff-<agent>-<id>.md) and copied to the clipboard
  4. Open the target product, start a new chat, paste, continue

Long conversations? Oversized packages auto-trim to a recap plus key turns, but the untrimmed transcript is exported alongside (-full.md) and named in the package header — a file-capable target agent (Claude Code, Codex, …) will read the full history itself and use the inline excerpt as a table of contents. Workspace-scoped sessions also carry their project path so the target agent reads the repo for code-layer context.

History — a cross-agent chat archive

History tab

Aggregates every agent's history into one neutral, searchable, backup-friendly archive. Click 聚合并生成时间线 (Aggregate & build timeline) and the output directory gains:

Documents/Myco/chat-archive/
├── index.json           # manifest
├── <session>.md/.json   # neutral format any tool can read
└── viewer.html          # single-file offline timeline — double-click to
                         # browse & search (follows OS dark/light, with a
                         # manual toggle that remembers your choice)

The archive contains your private conversations. It is git-ignored by default — review before sharing or committing anywhere.

Settings — theme & paths

Settings tab

  • Theme — dark / light toggle (the sun/moon button in the header does the same)
  • Script resources (read-only) — where the Python engine and bundled skills live
  • Output directory (writable) — handoff packages, archives and distributions all land here; defaults to Documents\Myco, one click to open
  • Python interpreter — the one actually in use (full distribution uses the bundled python\python.exe)
  • Re-detect — rescan agents after installing a new tool

Privacy promise

  • Every agent's local data is accessed read-only; SQLite databases are opened in read-only mode
  • Myco never writes back into any agent's storage
  • Handoff / archive operations only produce text files, all inside your own output directory
  • No network upload — everything happens on your machine

FAQ

Q: "Python not found"? The full distribution bundles Python, so this shouldn't happen. On the light package or from source, install Python 3 or point MYCO_PYTHON at an interpreter.

Q: An agent shows "结构已变" (layout changed)? That product moved or reshaped its session storage in an update. Myco reports it honestly instead of guessing. Please open an issue with the new path — it's usually a one-line engine/agents.json fix.

Q: Why is the session count approximate? For jsonl-based agents the count comes from counting files — fast but not parsed per file. The Relay list is parsed for real and exact.

Q: I can't see the glass effect? Acrylic shows whatever is behind the panel — over a plain background the blur is featureless by nature. Put a colorful window behind it and it pops. Windows 10 / older Win11 falls back to an opaque backdrop automatically.

Q: Can I skip the UI and use the CLI? Yes — the engine is a plain CLI (run inside the install directory; use python\python.exe for the bundled interpreter):

python engine\agent_status.py            # agent detection overview
python engine\sync_chats.py --dry-run    # preview history aggregation
python engine\handoff_chat.py --list     # list sessions available for handoff
python engine\distribute.py --dry-run    # preview skill distribution

Advanced: environment variables

Variable Effect
MYCO_REPO Override the resource root (point at a source tree during development)
MYCO_WORKDIR Override the output directory (default Documents\Myco)
MYCO_PYTHON Override the Python interpreter
MYCO_PREVIEW=1 Open the panel as a normal window (screenshots/demos); combine with MYCO_TAB