3dl-dev
OrganizationA tap of tested, self-building hoist skills for real apps.
Categories
Indexed Skills (11)
deploy
Set up and run honcho on this machine, then self-test it and report honestly what worked. It ships as a recipe, and on first use it fetches a verified toolchain, brings honcho up in a sandbox, and grades it.
develop
The dev-team lifecycle for a self-hosted open-source app. You (the agent) understand how to iterate on it and test it, fork it or keep a separate tree, pull from upstream (verified against the tests), and contribute back when the user chooses.
sysop
Take preflight's scoped plan and stand the app up on the target, then operate it. Composes external infra skills, owns secrets, keeps petard's corpus fresh. Deploys into a resolved isolation substrate (a same-host namespace, or an environmental sandbox where a deploy cannot reach host state), never the app's own.
petard
The no-frontier operational fallback. When the frontier model is down and credits are out, you still get the exact command to run, grounded in your system's own ground truth. petard composes arlo; it never invents a command.
preflight
Scope a deployment with the user and tell them early whether it will work, deploying nothing. Emits the scoped plan and a feasibility verdict for sysop.
arlo
Lights-out ops. A real, local operator: when the frontier model is down and credits are out, you type your own words and arlo hands back the exact command to run, grounded in your system's own ground truth. It never invents a command.
start
Set up arlo for this project while the frontier is up: harvest the project's real ground truth into capability cards, resolve the local-model tier, and wire the `arlo` command so lights-out ops work when the frontier is down. Run once per project. Also invocable as /arlo:setup.
update
Re-sync arlo with reality: re-harvest this project's ground truth (cards are a regenerated projection, never stored), pick up any arlo skill improvements, and report what drifted — the guard against a card that still says `--parent` after the flag became `--parent-id`. Run periodically and whenever the project's commands change.
upgrade
Pull the latest arlo from upstream (new rungs, sharper harvest rules, fixes), then reconcile each project with /arlo:update. This is how arlo-the-tool moves forward; /arlo:update only re-reads your project against the arlo you already have.
remove
Cleanly remove arlo from this project: take the `arlo` command off the PATH and delete the project's `.arlo/`. Asks before deleting the shared local-model runtime, which other projects may still use. Never touches the project's own ground truth.
setup
Alias for /arlo:start — set arlo up for this project (harvest ground truth, resolve the model tier, wire the `arlo` command). Follow the start skill.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.