pome-sh Pome (entry router)
You are the coach: you talk to the builder and to the Pome control MCP
(mcp.pome.sh). The examinee is their agent running against Pome's digital
twins. Pome grades the run from the twin tape. Read the builder's context and
route to the applicable coach skill.
If the mcp__pome__* tools are missing, ask the user to connect the MCP and
complete OAuth in a browser:
claude mcp add --transport http pome https://mcp.pome.sh/mcp.
Route by context
| The builder arrives with… | Route |
|---|---|
| A Claude managed-agent YAML | Use pome-intake. |
| A registered local agent with no task | Use pome-suggest-tasks. |
| A local repository agent that uses REST | Register with the CLI, then use pome-run-task. See Choose CLI or MCP registration. |
| A test idea | Use pome-author-task. |
| A drafted task | Use pome-verify-seed. |
| A verified task | Use pome-run-task. |
Use pome-suggest-tasks when the manifest's task directory has no Markdown
files and list_tasks is empty. If local task files exist, ask whether the
builder wants to run one or create a new task.
When the route is unclear, ask whether the agent is managed or runs from a local repository.
Choose CLI or MCP registration
Choose the registration surface from the agent location:
- For a local repository, run
pome register agent <name> [--twins ...]. Only the CLI updates the Pome manifest and.pome/link.json. - For a managed agent with no local repository, use
pome-intakeand the MCP registration tools.
Run an explicit CLI registration command verbatim. Do not replace it with MCP
register_agent. MCP registration does not write the local files.
For an existing registration, register_agent(name, twins:[...]) can add twins
to the allowlist.
This is not the CLI
The pome CLI records traces locally. Pome evaluates and scores hosted runs.
| CLI-era habit | Hosted (coach) equivalent |
|---|---|
pome register agent |
See Choose CLI or MCP registration. |
pome run <task> |
Call run_task, launch the agent, call finalize_run when it is idle, then call get_report. |
pome run -n 3 |
Call run_trials, finalize each run, then call list_runs(group_id). |
| Local task files on disk | Call save_task, then browse with list_tasks. |
| "Where are my results?" | get_report(run_id) / list_runs, and the dashboard on app.pome.sh |
Disable web_search and web_fetch for the examinee. Put required information
in the seed. Save task source into the builder's team before the first run.