Athena Skills lets a team turn a procedure that worked into a reusable agent skill. Athena Intelligence writes the steps from a session into a SKILL.md file in the open Agent Skills format, a person publishes the skill as a versioned release, and any teammate's agent run can load the published version. Earlier releases stay available for rollback.
Most teams have one session that went right: the quarterly deck built in the house template, every figure checked against the actuals. Next quarter someone else explains the procedure again from memory, and the deck comes out a little different. A skill keeps the steps that worked.
Athena Intelligence, founded in 2022, builds an agent platform for regulated enterprises that can be self-hosted, where people and agents build agents, workflows and applications together. This dispatch follows the launch video, where one analyst's Q3 deck becomes a skill. Figure 1 shows the idea on one page, and the Skills product page has more.
- A skill is a SKILL.md file with a name, a description and the steps an agent follows.
- Athena can write the skill from a chat that already did the work, and a person reviews the draft.
- Publishing makes a numbered release. Agent runs read the published release, and the live doc stays an editable draft.
- Every release is kept, and Roll back to this makes an older one current again.
Contents
01What problem do agent skills solve for teams?
Agent skills keep a working procedure from getting lost between sessions and between people. Without a skill, the steps behind a good result live in one chat history or in one person's head.
The usual workaround is a prompt pasted into a wiki page. Each person edits a copy, and the output drifts. When the person who knew the steps moves on, the steps go too: the median US wage and salary worker had been with their current employer for 4.1 years in January 2026, according to the Bureau of Labor Statistics. Figure 2 sets the two paths side by side.
More of this work now runs through agents. Gallup found that among US employees in remote-capable roles, 66% used AI at work and 40% used AI a few times a week or more in late 2025. Among EU enterprises that bought cloud services in 2025, Eurostat reports that 71.7% used cloud office software and 71.5% used cloud file storage, so the decks and sheets these procedures produce are shared files too.
02What is an agent skill?
An agent skill is a folder of instructions an agent loads when a task calls for the skill. The required file is SKILL.md: a short header with a name and a description, followed by the steps in plain Markdown.
The format is open and documented in the Agent Skills specification. Agents read each skill's name and description up front and load the full instructions when a task matches.
In Athena, a skill is an asset, created with Create Skill in the new-asset picker or by asking Athena in a chat. The editor has three fields, Skill name, Description and Instructions (SKILL.md), and a green banner reads “Valid SKILL.md — changes are saved automatically.” Figure 3 shows the skill from the video.
03How do Athena Skills work? One example, start to finish
The launch video runs three steps: capture, publish and load. Figure 4 lays out the steps.
In the demo, an analyst's chat builds the Q3 business review deck from Q3_actuals.xlsx in the team template, and a teammate, Priya, later uses the skill for Q4. All names, files and numbers are demo data.
- Capture. With the Skill toolkit @-mentioned, the analyst types “turn what we just did into a skill called quarterly-deck-review.” Athena calls Create Skill Asset, and the new SKILL.md lists four steps: read the actuals and last quarter's deck, pull revenue, gross margin and opex by region, check every figure against the source and cite the source on the slide, and build six slides.
- Publish. The Publish skill dialog takes a snapshot of the SKILL.md, and Releases shows v1 as Current. The analyst adds a fifth step, “Flag any figure that moved more than 5% vs. last quarter,” and publishes v2. The v1 row stays, with Roll back to this.
- Load. Priya opens a new chat with Q4_actuals.xlsx, picks quarterly-deck-review from the Skills list in the @ menu and asks for the Q4 business review. The first tool card reads /skill-assets/quarterly-deck-review/SKILL.md, then the deck builds with the same steps and template.
Table 1 compares a procedure shared by hand with the same procedure shared as a skill.
| Task | Shared by hand | Shared as a skill |
|---|---|---|
| Write the procedure down | A wiki page, if someone finds the time | Athena drafts the SKILL.md from the chat that did the work |
| Share the procedure | A prompt copied into each chat and edited | Pick the published skill from the @ menu or attach the skill to an agent |
| Change a step | A channel message, and everyone updates a copy | Edit the draft and publish a new release |
| See who changed what | Compare pasted prompts by hand | The Versions drawer lists the person behind each version |
| Undo a bad change | Search an old chat for the previous text | Roll back to this on an earlier release |
| Run the procedure next quarter | Explain the steps again | Mention the skill, and the run reads the published SKILL.md |
04How do versions and rollback work for agent skills?
Publishing a skill freezes the current SKILL.md as a numbered release, and agent runs read that release. The draft and the published version stay separate, so a half-finished edit stays out of everyone's runs.
A chip in the editor shows the state: Not published, Published or Unpublished changes. The note under the banner reads “Agent runs mount the published SKILL.md; the live doc stays your editable draft.” Each publish adds a row to Releases, the newest marked Current, and older releases keep a Roll back to this button. Figure 5 follows one skill through two releases.
Rolling back a skill changes which procedure future runs follow. For changes a run already made, the platform's audit trail and rollback applies, with irreversible actions flagged. Human approval can be required on any tool.
05Is Athena secure enough for regulated companies?
Yes. Athena Intelligence's security program includes SOC 2 Type II and HIPAA, third-party penetration tests and zero data retention agreements with leading model providers, and Athena does not train on customer data. Figure 6 shows where skills sit.
Skills live inside whichever deployment the company chose: a managed cloud, your own VPC on AWS, Azure or GCP, on-premises, or air-gapped, including AWS GovCloud. A Big Four audit firm has run Athena in production for over a year, in a 1,250-user deployment across five localities. The deploy anywhere and permissions and identity pages go into detail.
06How do we get started with agent skills?
Pick a procedure your team repeats and already does well. Run the procedure once in a chat until the output is right, @-mention the Skill toolkit, and ask Athena to save the work as a skill. Tighten the steps, publish, and ask one teammate to load the skill next time. Skill assets carry a Beta label in Athena today. To run a procedure on a schedule, see the AOPs page.
Curious how a skill would work on your team's procedures? We're happy to show you.
07Questions teams ask
What is an agent skill?
An agent skill is a reusable set of instructions that teaches an agent how to do one task the way a team does the task. The core is a SKILL.md file with a name, a description of when to use the skill, and the steps in Markdown. A skill can also carry scripts, reference documents and templates.
How do we keep everyone on the team using the same version?
Publish the skill. Publishing takes a snapshot of the current SKILL.md as a release, and agent runs read the published release, so everyone who loads the skill gets the same steps. Edits stay in the draft, marked Unpublished changes, until someone publishes again.
Can we roll back a skill?
Yes. Every release is kept in the Releases list, and each older release has a Roll back to this button. Rolling back changes which steps future runs follow. Changes a run already made are handled through Athena’s audit trail and rollback.
Can a skill written for one agent work in another?
Athena uses the open Agent Skills format, the same SKILL.md structure other agents that support the format read, so the steps are portable text. The tools and data a run can reach depend on where the skill runs, so test a skill on a real task in each place your team uses the skill.
How is a skill different from a prompt or an AOP?
A prompt is typed into one chat and stays there. A skill is a named procedure with numbered releases that any run can load. An agent operating procedure (AOP) is a workflow Athena runs on a schedule or when an event triggers the workflow.
Run this on a schedule
Start from a ready-made procedure instead of a blank chat.