name: hq-cowork-cli description: Run long-tail HQ CLI commands through guarded host-side Cowork tools. allowed-tools: mcp__hq__hq_cli, mcp__hq__hq_run, mcp__hq__hq_whoami
/hq-cowork-cli — Long-tail HQ from Cowork
Use this when Cowork needs an HQ capability that does not have a dedicated
hq-cowork-* skill yet.
Prefer dedicated tools first:
- Search:
/hq-cowork-search - Sync:
/hq-cowork-sync - Files/share:
/hq-cowork-files,/hq-cowork-share - Secrets:
/hq-cowork-secrets - Meetings/sources/signals:
/hq-cowork-meetings
mcp__hq__hq_cli
Runs hq <args...> on the host. Pass argv, not a shell string.
{
"args": ["sync", "status"],
"cwd": ".",
"timeoutMs": 60000
}
The tool blocks:
hq login,hq logout,hq onboard- browser/session auth flows other than
hq auth statusandhq auth refresh - secret-value output (
hq secrets env,hq secrets get --reveal) - raw
hq secrets set|execandhq runthrough the escape hatch
Use mcp__hq__hq_run for hq run.
mcp__hq__hq_run
Runs schema-driven commands with HQ secrets injected via .env.schema.
Secret values stay in the child process env; only command output returns.
{
"cwd": "repos/private/example-app",
"company": "indigo",
"schema": ".env.schema",
"cmd": ["npm", "test"],
"timeoutMs": 120000
}
For validation only:
{
"cwd": "repos/private/example-app",
"check": true
}
Rules
- Never ask the user to paste secrets inline.
- Never use the escape hatch for secret-value retrieval.
- Keep
cwdinside the HQ root. The MCP server enforces this. - For destructive or cloud-provisioning commands, explain the concrete effect before calling the tool.