Cloud agents

Skill files, and why a worker's capability should be readable

The difference between a prompt and a playbook is that one of them is a document your colleague can edit.

The short answer

A skill file is a markdown playbook that tells an AI worker how to do one kind of work: when the method applies, the steps to follow, and the shape of the output. Polaris stores each as a row with a name, a description and a markdown body, mirroring the SKILL.md format. Every skill attached to a worker is pasted into the system prompt in full at the start of each job.

Format: Name, description, markdown body
Attached via: A list of skill ids on the worker
Loaded: In full, at the start of every job

What a skill file is for

Instructions describe who a worker is. A skill describes how a particular job gets done. Splitting them means one worker can carry several methods, and one method can be shared by several workers without being retyped.

Polaris keeps skills in an org-wide library. A worker row holds a list of skill ids, and at job time the runtime fetches those rows and pastes each one into the system prompt under a heading with its name and description, followed by its complete body. There is no summarisation step and no retrieval ranking: an attached skill is present in full, every time.

How a skill reaches the model

1
Someone writes it
A name, a one-line description of when it applies, and a markdown body. The shape that works is a when-to-use section, a numbered method, and an output section describing the deliverable.
2
It joins the org library
Skills are org-scoped rows readable and writable by members, so a method one person refines is available to every worker in the organisation.
3
It gets attached to a worker
The worker's row carries an array of skill ids. Attaching and detaching is an edit to that array, not a rewrite of the worker.
4
A job compiles it in
At claim time the runtime reads every attached skill and renders them under a Your skills heading in the system prompt, above the operating rules for the session.
5
The worker follows it
The method in the file shapes the session directly, because it is sitting in the same prompt as the task brief and the acceptance criteria.

The starter library, as shipped

Six skills are seeded into every organisation. These are the real names and the real intent.

web-research
Decompose a question into three to five queries first, prefer primary sources, cross-check load-bearing numbers across at least two, stop when new sources stop changing the answer, and end with a sources list that only cites what was opened.
competitive-analysis
Build and confirm the competitor set before going deep, then capture offering, pricing model, positioning, one strength and one exploitable weakness for each, normalised into one table. Facts are separated from the analyst's read.
summarization
Identify the reader and the decision before writing, extract claims and numbers exactly rather than paraphrasing figures, and cut anything nobody would act on.
copywriting
Restate the audience and the single action, draft three angles with a marked recommendation, keep sentences short and concrete, and cut filler.
content-strategy
Anchor on the business goal, propose three to five pillars with the audience question each answers, map to a realistic cadence, and flag what should not be published.
social-posts
Write three hook options first, respect platform norms without being generic, one idea per post, and hand back the rejected hooks labelled so a human can swap them.

Hidden prompt versus readable playbook

A prompt in someone's config

Lives on one machine, in one person's setup
Improvements are invisible to everyone else
Nobody can review why the output has this shape
Leaves with the person who wrote it

A skill in the library

A row any member can open and edit
One edit changes every worker carrying it
The method behind an output is inspectable
Survives the person who wrote it

Assigning a task to an AI worker

There is no prompt box. The task is the prompt, and the checklist is the contract.

How an AI worker gets tool access

The interesting part of tool access is not the list. It is where the credential lives and who can read it.

Skill file

The capability an agent has, written down where a person can read and edit it.

AI worker

The difference between an agent you prompt and an agent you assign work to.

Hire an AI research analyst

Assign the question on Tuesday afternoon and read the brief before your Wednesday call, with every claim carrying the URL it came from.

Questions people ask

+How is a skill different from the worker's instructions?

Instructions define the worker as a whole: its name, mission and standing rules, such as never marking a task done. A skill defines a method for one kind of work and can be attached to several workers. Both end up in the same system prompt, with instructions first and skills underneath.

+Can I edit a skill file?

Yes. Skills are org-scoped rows that any member of the organisation can read and edit, and the change applies to every worker carrying that skill on its next job. There is no hidden or vendor-locked copy of the text.

+Do skills consume context on every job?

Yes, because attached skills are pasted in full into the system prompt at the start of each session. That is a deliberate trade: a worker carrying three focused playbooks behaves more predictably than one carrying ten sprawling ones, so attach the methods that job actually needs.

+Can a worker gain a skill during a session?

No. The skill set is compiled once, when the job is claimed. Adding a skill applies from the next session onward, which keeps a running session's behaviour reproducible from its inputs.

Cloud agents that keep working after you close the laptop

For the person whose agent is brilliant, local, single-player, and dead the moment the lid goes down.

Assigning a task to an AI worker

There is no prompt box. The task is the prompt, and the checklist is the contract.

How an AI worker gets tool access

The interesting part of tool access is not the list. It is where the credential lives and who can read it.

Turning a single-player agent into a team member

The change is not a better prompt. It is putting the worker in the same table as the people.