dstack skill
A skill set for website performance and scalability, currently holding ooda-loop: observe, orient, decide, act.
- 135 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill is meant for production metrics and cloud services.
- After your approval the agent changes the environment.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/dstackDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill is meant for production metrics and cloud services.
- After your approval the agent changes the environment.
This is third-party code. Review the repository files before installing.
What it does
dstack gathers the author daily skills for website speed and availability. Right now the repository holds one skill, `ooda-loop`, for latency, throughput, errors, cloud spend and outages. The agent follows a loop: observe, orient, decide, act, then observe again. The first three phases are read-only, with no edits, deploys or restarts. Before starting the agent pins a measurable goal, such as "p95 under 300 ms on /checkout". At the decision it proposes one small reversible action with a hypothesis, abort condition, impact assessment and rollback, and waits for explicit approval. After acting it compares metrics before and after and writes a cycle report.
Who it is for. Engineers and SREs who debug a slow or unstable site with an agent and want changes to happen only after their approval.
Good fit when
- p95 latency grew and the cause is sought in metrics, not guesses.
- An outage is under way and any mitigation must pass through your approval.
- The cloud bill grew and spend should drop without breaking the SLO.
- The agent has observability MCPs and CLIs connected and should work in a loop.
Not a fit when
- There is no access to live metrics, logs or traces: the loop stops at observation.
- You need a general engineering workflow set: the author recommends another package for that.
- Changes must go in without a pause for approval.
Example request
Run ooda-loop with the goal of p95 on /checkout under 300 ms. Start with observation and stop before any action.Limitations
The author marks the repository as work in progress, and it holds one skill. Without access to real signals (MCPs, cloud vendor CLIs, logs) the agent asks for access and stops. The write guard does not replace your access policies: give the agent read-only rights until an action is approved. The loop relies on live data, so work out which observability systems are connected before the first run.
How to disable. Remove the `ooda-loop` skill from the project or user skills directory (`npx skills remove` per the tool docs). Revoke connected MCPs and keys separately.
FAQ
What is in the set?
Currently one skill, `ooda-loop`. List the available ones with `npx skills add daniellockyer/dstack --list`.
When does the agent make changes?
Only in the act phase and after explicit approval. Load tests, cache warming and toggling log levels also count as acts.
How do I install globally?
Add the `--global` flag to the install command: `npx skills add daniellockyer/dstack --global`.
Where does the agent get data?
From MCPs, vendor CLIs and APIs, observability tools and local logs. Inventing numbers is forbidden, and without data it asks for access by name.
Is it suitable for general development?
No. The skills focus on performance and scalability, and the author recommends the pstack package for general engineering workflows.
Discussion
No questions or reviews yet.
Related
SkillsThe all-deploy skill puts a site, app or API online: audit, preview, response check and rollback
A pattern library and skill set for recurring agent loops: issue triage, dependency cleanup, PR babysitting
Skill configures OpenTelemetry: collectors, sampling, security and telemetry from AI agents, without flooding the context.
Official Dynatrace skills for agents: DQL, traces, logs, Kubernetes, cloud and security work inside Claude Code and Cursor.
Official Datadog skills: monitors, logs, APM, cloud integrations, RUM and LLM observability from inside the agent
KubeShark skill: the agent finds the failure mode first, then writes manifests, Helm and Kustomize to official practice with a rollback plan
README in short
The purpose of the set, a work-in-progress note, a pointer to a broader package, install via `npx skills` and the MIT license.