intendant-remote-compute

Offloads heavy builds, tests, and clippy sweeps to Intendant's remote-compute lane.

14|Updated Mar 2, 2026
One-click install
npx skills add https://github.com/intendant-dev/Intendant --skill intendant-remote-compute-intendant-dev
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: intendant-remote-compute
Source: https://github.com/intendant-dev/Intendant/tree/main/plugins/codex-cloud-remote-compute/skills/intendant-remote-compute
Command: npx skills add https://github.com/intendant-dev/Intendant --skill intendant-remote-compute-intendant-dev

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running full workspace compiles, broad test suites, clippy sweeps, and benchmarks on a local machine creates contention, crash pressure, and long waits during agent-driven development sessions. This Skill routes that heavy, platform-neutral work through the Intendant daemon's provider-neutral remote-compute lane instead of loading the local machine. ## Core Features & Use Cases - Remote job lifecycle via CLI: Submit, poll, wait on, and cancel remote jobs with intendant ctl remote start|status|wait|cancel, using real flags instead of hand-built JSON, from any supervised or unsupervised session. - Durable compiler caching: Request --cache durable_sccache to reuse Rust compile outputs across worker replacements, namespaced by the supervised session's project root. - Honest failure handling: Never silently fall back to heavy local work when the lane fails; report the acquisition stage, provider status, and task URL, then continue only with cheap local steps. - Use Case: During a coding session you need to run cargo clippy --workspace -- -D warnings. Instead of running it locally, you push your branch, run "$INTENDANT" ctl remote start --branch my-branch --revision <sha> -- cargo clippy --workspace -- -D warnings, then block with ctl remote wait until the job finishes. ## Quick Start Ask the agent to offload the full workspace clippy run to the remote-compute lane using intendant ctl remote start with your pushed branch and revision, then wait for the job to finish.

Frequently Asked Questions about intendant-remote-compute

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a heavy build remotely with intendant ctl remote?▼

Push your branch, then run intendant ctl remote start --branch my-branch --revision <sha> followed by the command, such as cargo clippy --workspace. The start verb returns a job id immediately, and ctl remote wait JOB --for SECONDS blocks until the job reaches a terminal state.

What work should stay local instead of using remote compute?▼

Cheap commands like focused unit tests, single-crate checks, quick greps, and small platform-specific checks stay local. Test suites asserting non-root permission or sandbox behavior also stay local because remote workers run as root and those tests fail there.

Does durable_sccache caching work from an unsupervised shell?▼

No. The durable_sccache relay namespaces its cache repository by the supervised session's project root, so unsupervised calls fail early with an error about requiring a supervised project root. Omit the cache flag and accept a cold build, or run from a supervised session.

Why is my remote job stuck in the acquiring state?▼

An acquiring job may still be creating a cold worker, which can take tens of minutes; automatic acquisition allows one hour by default. Check job.acquisition via ctl remote status for the stage, provider task URL, and last provider error instead of resubmitting a duplicate.

What should I do when the remote compute lane fails?▼

Report the acquisition stage, task URL, provider status, and last provider error, then continue only with genuinely cheap local steps. Never silently run the expensive work locally; a signed-out failure means the provider lease died and must be re-armed from the Vault tab.