Make third-party models work inside Claude Code as close to Opus as the plugin layer can drive them.
hello2cc does not replace your model gateway, provider mapping, or account permissions.
Its job is simpler:
If you already connected GPT, Kimi, DeepSeek, Gemini, Qwen, or other third-party models to Claude Code,
hello2cckeeps pushing them toward Opus-compatible Claude Code behavior: native capability priorities, tool and agent choice, team/task workflow use, failure recovery, and output style.
Language: English | 简体中文
Compared with 0.4.9, this release changes how hello2cc coexists with surfaced host skills and workflows:
| 0.5.0 note | What you should notice |
|---|---|
| Visible host skills/workflows can own the main flow | Plugins such as superpowers are less likely to be overridden by a parallel hello2cc playbook |
| Native output style still stays on | Even when a host workflow leads, replies remain closer to Claude Code / Opus-style presentation |
| Tool semantics and protocol adaptation remain active | Third-party models still benefit from native tool-use hints, parameter cleanup, and failure debounce |
| Session-start injected bootstrap skill docs are surfaced | using-superpowers-style bootstrap content can be recognized as a real host skill surface |
| Common problem | What hello2cc improves |
|---|---|
| A matching skill or workflow already exists, but the model keeps rewriting the process | Encourages the model to continue with the surfaced or already-loaded workflow |
| The session already exposes MCP resources or tools, but the model still takes a detour | Nudges the model toward the most direct path first |
| Plain parallel workers get confused with team or teammate semantics | Reduces avoidable agent routing mistakes |
| The model can answer, but does not pick the right Claude Code capability | Pushes tool, agent, workflow, and MCP choice back toward native Claude Code priorities |
| The model depends too much on wording or keyword hints | Switches to language-agnostic semantic matching inside host-exposed candidate boundaries |
| The output style drifts away from native Opus | Forces a tighter Claude Code-native result style |
| The model drifts into verbose meta narration | Keeps responses closer to concise native Claude Code style |
- You already use Claude Code with third-party models through CCSwitch or another mapping layer
- You want those models to behave more like a native Claude Code session
- You have skills, workflows, MCP servers, or plugins installed and want them to be used more reliably
- You want parallel agent work to choose a more appropriate path
- Setting up provider accounts, API keys, or gateway access
- Exposing tools that Claude Code did not expose in the first place
- Replacing CCSwitch
- Overriding higher-priority repository rules such as
CLAUDE.md,AGENTS.md, or direct user instructions
| Item | Value |
|---|---|
| Install flow | 3 steps |
| Extra command required after install | 0 |
| Common config profiles | 2 |
| Main goal | 1 — make third-party models behave more like native Claude Code / Opus sessions |
|
Better continuity when a workflow is already visible or already in progress. |
Stronger preference for capabilities that are already available in the session. |
|
Keeps one-shot parallel work on ordinary agent paths, while sustained collaboration is more likely to use a real team + task-board flow. |
Less unnecessary meta narration and fewer avoidable routing mistakes. |
git clone https://github.com/hellowind777/hello2cc.git
cd hello2ccclaude plugins marketplace add "<repo-path>"Replace <repo-path> with your local hello2cc repository path.
claude plugins install hello2cc@hello2cc-localThen reopen Claude Code or run /reload-plugins.
- No extra manual entry point is required
- Installing the plugin does not write
agent=hello2cc:nativeinto Claude Code settings - Plugin enablement stays under Claude Code's own plugin state, not a plugin-shipped
settings.json - Third-party models are more likely to use session-visible capabilities directly
- Ordinary parallel agents are less likely to be misrouted
- Team/subagent sessions are less likely to suffer from oversized injected context
Good when your model mapping is already handled by CCSwitch and you want hello2cc to stay on its default strong alignment path.
{
"mirror_session_model": true
}Good when you want most agents to use the same Claude slot.
{
"mirror_session_model": true,
"default_agent_model": "opus"
}If your real target model is mapped through CCSwitch, keep the actual mapping there.
In hello2cc, prefer stable Claude slot values such as inherit, opus, sonnet, or haiku.
- Keeps hello2cc's native output shell active without forcing a private workflow
- Lets surfaced host skills/workflows lead the main flow when they are the better owner
- Continues native-style tool semantics, protocol cleanup, and repeated-failure debounce
- Improves coexistence with session-start bootstrap skills and workflow plugins
flowchart LR
A[Third-party model already connected in Claude Code] --> B[Open a normal Claude Code session]
B --> C[Session exposes tools, agents, skills, workflows, or MCP]
C --> D[hello2cc helps the model choose the path more naturally]
D --> E[Fewer detours and cleaner agent routing]
style A fill:#e3f2fd
style B fill:#e3f2fd
style C fill:#fff3e0
style D fill:#e8f5e9
style E fill:#4caf50,color:#fff
If you changed the local repository or want a clean reinstall:
claude plugins uninstall --scope user hello2cc@hello2cc-local
claude plugins marketplace remove hello2cc-local
claude plugins marketplace add "<repo-path>"
claude plugins install hello2cc@hello2cc-localThen reopen Claude Code or run /reload-plugins.
Try these in order:
- Reopen Claude Code or run
/reload-plugins - Confirm the plugin is installed and enabled
- If you upgraded from a local clone, reinstall it cleanly
That banner can persist in the current or resumed session because Claude Code keeps session agent state separately from plugin enablement.
Current hello2cc no longer ships a plugin-side settings.json to force agent=hello2cc:native, so a clean reinstall or a fresh session stops new unintended agent injection.
Check whether:
- That capability is actually exposed in the current session
- A higher-priority project rule or user instruction is restricting it
- You are continuing the same workflow instead of starting a different one
- If a plugin injects a bootstrap skill at session start, update to
0.5.0or later so hello2cc can surface that skill boundary instead of competing with it
From 0.5.0, hello2cc keeps its output-style shell and tool semantics active, but it no longer needs to own every main-thread workflow.
If another plugin surfaces a real skill/workflow owner, hello2cc now yields the main flow more cleanly instead of fighting it with a parallel execution playbook.
Update to the latest version, reload the session, and reinstall if needed.
Recent versions add a compatibility layer for plain-text SendMessage.
Update to 0.5.0, then reload the plugin cleanly.
Current versions remove the plugin-side pre-deny for explicit teammate retries, so missing-team handling falls back to Claude Code's native team error path instead.
Update to 0.4.6 or later, then reload the plugin cleanly.
Recent versions tighten short-query shaping and treat Did 0 searches as an empty search attempt instead of a successful result.
Does hello2cc replace CCSwitch?
No. Keep model mapping in CCSwitch. hello2cc focuses on what happens after the model is already running inside Claude Code.
Does it enable tools that Claude Code did not expose?
No. It only helps the model use capabilities that are already available in the current session.
Do I need to switch an output style manually?
Usually no. After installation, it should work without an extra manual entry point.
Will installation force-select hello2cc:native for the current thread?
No. The plugin ships the native agent as an available option, but it no longer injects a default agent setting into Claude Code.
Will it block my existing skills, plugins, or MCP servers?
No. The goal is the opposite: make those existing capabilities easier for third-party models to discover and use.
Does every multi-agent task become a team?
No. One-shot parallel work can stay on ordinary agent paths. More sustained collaboration is more likely to use a team flow.
Do I need to set a default agent model?
Only if you want a stable default Claude slot for most agents. If your mapping is already managed well elsewhere, you can keep the config minimal.
Can I still switch hello2cc into a lightweight compatibility-only mode?
No. The current direction keeps hello2cc as a default strong alignment layer instead of falling back to a thin sanitize-only shim.
- Issues: https://github.com/hellowind777/hello2cc/issues
- Releases: https://github.com/hellowind777/hello2cc/releases
Apache-2.0