Skip to content
Get packs

Guide · Agent skill

How to use the Prioritize agent skill: setup and examples

Updated · written with AI from the project's own documentation and checked against our catalogue data

Prioritize is an agent skill from the LifeOS personal system. It gathers what your system is currently seeing, such as open work items, graded bookmarks and stories, new notes, open upgrade records and your TELOS goals, and turns it into a short ranked list of what to work on next. This guide explains what the skill does, how to install it, how to ask for it in plain language, and what to check before you act on its suggestions.

Prioritize Recommendationsdanielmiessler/LifeOS · MIT

Quick start

  1. 1Check your LifeOS setup. Confirm that Config.json exists at LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Prioritize/ and that Bun is installed.
  2. 2Install the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
  3. 3Add your OpenAI key. Provide an OpenAI API key for the model step, following the Setup.md guide included with the skill.
  4. 4Schedule the refresh. Set a scheduler of your choice to run Prioritize.ts run every 120 minutes.
  5. 5Ask for your priorities. Ask the agent 'what should I work on next?' to see the focus summary and top three.
  6. 6Review the evidence. Ask why an item is recommended and read the linked inputs before you act on it.

What does the Prioritize skill make the agent do?

Prioritize is a recommendation engine. A collector reads local copies of your sources, fingerprints them, and skips the run if nothing has changed. When something has changed, it makes one low-cost OpenAI call that returns a short focus summary and ranked recommendations.

Each recommendation names a suggested next action and cites the real inputs behind it. Anything the engine cannot ground in the collected data is dropped. The results are shown on the Pulse /prioritize page, and the agent can run the skill on demand and explain or act on what it shows.

  • Sources: open work queue, Synapse-graded bookmarks and stories, new Cortex notes, open upgrade records, TELOS goals
  • Output: a focus summary, a ranked list, a next action and evidence for each item
  • Refresh: the page is refreshed about every 2 hours by a scheduled job
  • It recommends and explains; it does not create, edit or send anything itself

How do you install the Prioritize skill?

You can install the skill by downloading the ZIP from the skill's page on this directory and unzipping it into one of two locations: your user skills folder at ~/.claude/skills/, or a project's .claude/skills/ folder. The folder should end up as a Prioritize directory containing SKILL.md, Setup.md, Intents/ and Tools/.

If you use the Claude apps instead, upload the skill under Settings, then Capabilities, then Skills.

  • Option A: unzip into ~/.claude/skills/ for all projects
  • Option B: unzip into a project's .claude/skills/ folder for that project only
  • Option C: upload the skill in the Claude apps under Settings → Capabilities → Skills

What do you need before the skill works?

The skill is for people running a configured LifeOS install. Without a Config.json file, every command exits with an error that points to the setup guide. The skill also needs a Bun runtime, because its main tool is a Bun script, and an OpenAI API key, which is used for the model step.

The Pulse server at localhost:31337 is optional. It supplies Synapse data and the page itself. If Pulse is down, the run continues on the other sources and records the error. A scheduler of your choice is needed if you want the refresh to run every 120 minutes.

  • Bun runtime
  • A LifeOS install with Config.json at LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Prioritize/
  • An OpenAI API key
  • Pulse server at localhost:31337 (optional)
  • A scheduler set to run every 120 minutes
bun ~/.claude/skills/Prioritize/Tools/Prioritize.ts show --json

How do you trigger Prioritize with natural requests?

You do not need exact commands. The skill responds to plain requests such as asking what to work on, asking for the current priorities, or asking why something was recommended. The agent then either runs a refresh or reads the current results, depending on whether the stored results are stale.

If the latest results are older than the job cadence, the agent runs a refresh first. When you ask about a specific recommendation, the agent reads its evidence from the linked inputs.

  • 'What should I work on next?'
  • 'What's most important right now?'
  • 'Show priorities' or 'what is the system seeing?'
  • 'Refresh priorities' or 'run prioritize'
  • 'Why is this recommended?'
  • 'Let's do the first one' or 'dismiss that'

What does a typical run look like?

When the agent runs a refresh, it reports the focus summary and the top three recommendations first, then names any source errors. Each item shows its next action and the evidence behind it. If the inputs have not changed since the last run, the run is skipped with a 'no new inputs' result, which is a correct outcome and costs no model call.

To read the current results without refreshing, the agent uses the show command. A forced re-rank is only used when you ask for one.

  • Unchanged inputs: skipped, no model call
  • Changed inputs: one lightweight OpenAI call
  • Source failures: named in the output, not hidden
  • Dismissed items: recorded so they do not keep coming back
bun ~/.claude/skills/Prioritize/Tools/Prioritize.ts digest

What can you do with the recommendations?

When you pick a recommendation, the agent routes its next action to the skill that owns that kind of item, under your normal permission rules. For example, a work issue goes to the work skill, an upgrade record goes to the Upgrade skill, and a goal change goes to the Telos skill. A category with no route is offered to you as a plain next step.

Treat every recommendation as a suggestion. Your pick is what authorises the action, not the ranking.

  • Pick the first item: the agent routes it to the owning skill
  • Dismiss an item: recorded through the Pulse dismiss endpoint
  • Ask why: the agent reads the linked evidence
  • Ask for a re-rank: the agent may force a run only when asked

What safety points should you know?

The skill runs scripts and uses the network. The collector reads local mirrors, and the model call goes to OpenAI through an API key you provide. Pulse endpoints are used at PULSE_URL, which defaults to http://localhost:31337.

Labels in the results come from external content, so the agent treats them as data, not instructions. The engine never creates issues, edits TELOS or sends messages.

  • Runs scripts: yes
  • Uses network: yes (OpenAI call; Pulse at localhost by default)
  • Read-only inputs; display-only output
  • Review the evidence before acting, because whether a recommendation is worth doing needs your judgment

What people use it for

  • Starting the day with a short 'what matters most' view drawn from your open work and goals
  • Checking whether a bookmark or story you saved deserves action this week
  • Asking the agent to explain why an upgrade record was ranked high, then deciding whether to act on it
  • Reviewing stale or failed sources when the Pulse page shows an old or incomplete list

Frequently asked questions

Is the Prioritize skill free?

The skill is published under the MIT licence. The model step makes a paid OpenAI call with your own API key, so you pay OpenAI for those calls. Unchanged inputs cost no model call.

What does the Prioritize skill require?

It requires the Bun runtime, a LifeOS install with a Config.json file at LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Prioritize/, and an OpenAI API key. The Pulse server at localhost:31337 is optional, and a scheduler is needed for the 120-minute refresh.

How do I install the Prioritize skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.

Why does Prioritize say there are no new inputs?

The collector fingerprints its sources and skips the run when nothing has changed. This is expected behaviour and means no model call was made. Ask for a re-rank if you want a fresh run anyway.

Does Prioritize create tasks or edit my goals?

No. It is display-only and recommends and explains. It does not create issues, edit TELOS or send messages. Other skills handle those actions when you choose one.

Can Prioritize run without LifeOS?

No. Without a configured LifeOS install, every command exits because Config.json is missing. It is designed for people running the LifeOS personal system.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Prioritize suits people already running the LifeOS personal system who want a daily view of what matters most, with evidence attached to every item. Skip it if you do not have a LifeOS install, want a tool that manages tasks or edits goals for you, or plan to act on its ranking without checking the evidence.

Download the skill