Advanced workflows and cloud collaboration
Choose Advisor, Agent View, /goal, dynamic workflows, routines, /loop, Ultraplan and Ultrareview with caching and Passion8 boundaries in mind.
Advanced workflows combine local orchestration and official cloud collaboration. Local work can use Passion8 and your filesystem; cloud features generally depend on claude.ai, GitHub connections and Anthropic environments.
For subagents, background sessions, teams and channels, start with parallel agents and Agent Teams and Channels.
#Choose a capability
| Capability | Execution | Suitable for | Passion8 boundary |
|---|---|---|---|
| Advisor | Consults another model in the current session | Decisions, failure analysis, completion checks | Depends on gateway model support |
| Agent View | Local background session list | Several independent tasks | Local gateway configuration applies |
/goal | Automatic continuation in the current session | Long tasks with clear completion criteria | Local configuration applies |
| Dynamic workflows | Scripts orchestrating local subagents | Migrations, auditing, research | Local configuration applies |
/loop | Timed turns in the current session | Polling, CI waiting, maintenance | Local configuration applies |
| Routines | Anthropic cloud | Schedule/API/GitHub triggers | Usually does not inherit the local token |
| Ultraplan | Cloud planning, local/cloud execution | Research before a large implementation | Official environment during planning |
| Ultrareview | Cloud multi-agent review | Deep pre-merge bug finding | Official billing and limits |
#Advisor
Advisor consults another model at important points without switching the primary model for the entire conversation.
| Action | Command/configuration |
|---|---|
| Enable in a session | /advisor |
| Enable in settings | advisorModel |
| Enable at startup | --advisor |
| Disable | /advisor off or remove the setting |
| Scenario | Value |
|---|---|
| Major refactoring plan | Consult a stronger model about important disagreements |
| Repeated test failures | Reassess the approach |
| Completion review | Independently check omissions, risks and verification |
| Lower-cost primary model | Use Haiku/Sonnet routinely, Opus/Fable when needed |
| Item | Cache/cost effect |
|---|---|
| Advisor call | Additional model usage |
| Main prefix | Usually no complete cache rebuild required |
| Different advisor model | Separately billed/cached requests |
| High-risk completion check | Extra cost may be justified |
#Agent View
Manage several local background sessions in one panel.
| State | Meaning |
|---|---|
| Working | Work in progress |
| Needs input | Waiting for an answer or permission |
| Idle | Ready for more input |
| Completed | Finished successfully |
| Failed | Ended with an error |
| Stopped | Was stopped |
claude agents
claude agents --cwd ~/work/repo
claude agents --json
claude attach <id>
claude logs <id>
claude stop <id>
claude respawn <id>Use it for independent tasks such as a bug fix, tests and documentation review. Isolate interdependent file changes with worktrees instead of running them concurrently in one workspace.
#/goal
Claude checks the goal at the end of each turn and continues if it remains unmet.
| Mechanism | Next turn | Stop condition |
|---|---|---|
/goal | Immediately after the prior turn | Model confirms the criteria |
/loop | At the interval | User stops it or Claude judges completion |
| Stop hook | After the prior turn | Script/prompt decision |
Use specific goals:
/goal Continue fixing until npm run lint, npm run typecheck and npm run build pass, and the new pages have no broken internal links.Avoid vague ones:
/goal Make it good#Dynamic workflows
Scripts orchestrate many subagents for dozens or hundreds of tasks whose results need cross-checking.
| Property | Subagents | Workflows |
|---|---|---|
| Decisions | Current Claude turn | Script |
| Intermediate state | Conversation context | Script variables |
| Scale | A few workers | Potentially many agents |
| Repeatability | Prompt-dependent | Reusable saved script |
| Interruption | Restart current turn | Resume the same run |
| Bundled command | Purpose |
|---|---|
/deep-research <question> | Multiple perspectives, cross-checking and a cited report |
| Limit | Reason |
|---|---|
| No direct shell/file access in the workflow | Agents perform operations |
| No ordinary user input mid-run | Only agent permission prompts |
| Concurrency cap | Protect local resources |
| Total agent cap | Prevent runaway execution |
#Routines and /loop
| Method | Runs in | Minimum interval | Suitable for |
|---|---|---|---|
| Cloud routines | Anthropic cloud | One hour | Reliable schedules, APIs and GitHub events |
| Desktop scheduled tasks | Local Desktop | One minute | Local files and tools |
/loop | Current CLI session | One minute | CI, deployments and temporary maintenance |
/loop 5m Check whether CI passed. If not, summarize the failures and try to fix them.Cron example:
0 9 * * 1-5Cloud routines are not extensions of your local Passion8 session. Configure provider, secrets, network and repository permissions separately in their official cloud environment.
#Ultraplan
Research and planning happen in the cloud before you choose where to execute.
| Stage | Description |
|---|---|
| CLI start | Launch the plan locally |
| Cloud research | Browser session investigates code/task |
| Browser review | Edit and approve the plan |
| Cloud execution | Continue on the web |
| Return to terminal | Execute the plan locally |
This can reduce mistaken changes in large tasks. Cloud stages generally do not inherit local gateway settings.
#Ultrareview
Remote multi-agent review provides a deeper check.
| Property | /code-review | /review <pr> | /code-review ultra |
|---|---|---|---|
| Target | Current diff | GitHub PR | Diff or PR |
| Execution | Local session | Local session | Cloud sandbox |
| Depth | Quick feedback | Moderate PR review | Multi-agent verification |
| Duration | Seconds to minutes | Minutes | Typically 5–10 minutes |
| Cost | Normal usage | Normal usage | May use usage credits |
Use it for large pre-merge changes, security-sensitive work or issues requiring independent verification. Local review is faster for small changes.
#Cache strategy
| Workflow | Five-minute/one-hour guidance |
|---|---|
| Advisor | Separate usage; stable main prefixes can still hit |
| Multiple Agent View sessions | Independent prefixes per session |
/goal | Continuous turns; can benefit from one-hour TTL |
| Dynamic workflows | Independent subagent contexts; control fanout cost |
/loop | Stable prefixes and intervals under one hour can help |
| Cloud routines | Official cloud rules; no local cache sharing |
| Ultraplan/Ultrareview | Official cloud billing/cache policy |
#Official references
Support
Need help?
For setup, billing, or model issues, email us. Check the status page for uptime.
WeChat / QQ support is available at the bottom right.
