New in After Effects Beta: After Effects AI Assistant

Hey all! A quick update from our end: we’ve changed how the skills folder works, so the instructions in the original announcement for user-provided skills have been updated. Your skills folder will now arrive empty, so please feel free to save the below Readme, Template, and sample skills to your folder manually.

We will be saving these skills to Help pages soon for future reference/access.

Readme:

​
# After Effects skills

Starter skills for the After Effects Assistant. A skill is a small set of instructions that shapes how the Assistant works when you call it with `/skill-name`. It applies only to that skill's task — other skills are unaffected.

You don't need coding experience to use these files. Each skill includes default settings you can use without changes, plus a clearly marked place to make it your own. Open a `SKILL.md` and tell the Assistant your preferences, standards, and boundaries in plain language.

## How a skill is organized

Each skill lives in its own folder with one required file, `SKILL.md`:

```text
skills/
|-- preflight/
|   `-- SKILL.md
|-- spellcheck/
|   `-- SKILL.md
`-- my-skill/
    `-- SKILL.md
```

The folder name, the `name` in the file settings at the beginning of `SKILL.md`, and the `/skill-name` command all match. For a folder named `my-skill`:

```yaml
---
name: my-skill
description: Improve something in a clear, reusable way.
---
```

- **`name`** is the command you type after `/`.
- **`description`** helps the Assistant recognize when a request belongs to the skill. Write it the way you'd naturally ask for the task, rather than in technical terms.

## Included skills

- `/preflight` — check a project before rendering or delivery
- `/spellcheck` — proofread and correct text in After Effects

## Customize a skill

Every skill has two layers:

- **Default behavior** — what the skill does before you customize it.
- **Your preferences** — the rules you add to fit your workflow. When the two conflict, your preferences win.

To customize, open a skill's `SKILL.md` and find the section marked **"Customize this skill here."** Add short, specific rules, for example:

- "Use Canadian English."
- "Prefer subtle overshoot."
- "Never delete original layers automatically."
- "Keep our approved layer prefixes."

Each skill includes a few example preferences to show what works well — replace them with your own. You don't need to rewrite the whole file; start by editing the examples.

Skills are written in plain language on purpose: they describe the result you want and the choices the Assistant should make, not the mechanics of operating After Effects. The Assistant already knows how to use After Effects, so "always leave a 24 fps master" or "never rename my control layers" belongs in a skill; step-by-step panel instructions do not.

When you're ready for more, you can edit beyond the preferences. The sections used in these skills — What it does, Customize, Default behavior, Approach, Boundaries, Verify, Report — are a starting point, not a rule. Add, remove, or reorder sections to fit. A short, clear skill works better than a long, vague one.

## Create a skill

Start from `SKILL-TEMPLATE.md`, or copy an existing skill folder:

1. Create a folder for the skill.
2. Add `SKILL.md` inside it.
3. Set `name` in the file settings at the beginning of `SKILL.md` to match the folder name. This becomes the `/` command.
4. Write a `description` covering what the skill does and the requests that should trigger it.
5. Add your preferences in the Customize section.
6. Keep the defaults, boundaries, and examples that fit; remove the rest.

## Keep skills focused

A skill works best with one clear job. A project-organizing skill should describe how you want projects organized, not also try to be a general animation or troubleshooting skill. Keep unrelated tasks in their own skills.

Skills Template:

Skill Template:

<!– HOW TO USE THIS TEMPLATE

A skill lives in its own folder, and the instruction file must be named SKILL.md.

my-skill/
`-- SKILL.md

Keep three things the same:

the folder name (my-skill)
the frontmatter “name” (name: my-skill)
the slash command (/my-skill)
This template is written as a small, complete example skill you can copy and change. Replace the parts that describe the example with your own task. –>

name: my-skill description: Describe what this skill does and the kinds of requests that should use it. Write it the way a user would ask for the task, not in terms of After Effects internals.
My Skill
One sentence describing the result this skill should achieve.

Customize this skill here
This is where you make the skill your own. Add your preferences, rules, standards, or exceptions below. The Assistant follows these when the skill runs, and everything under “Default behavior” is what happens if you leave this empty. When your preferences conflict with the defaults, your preferences win.

<!– Good preferences are short and specific. You can add a preference, a rule, a standard, or an exception. You do not need to rewrite the rest of the skill.

Replace the examples below with your own. –>

Prefer subtle results over dramatic ones.
Never change anything I did not ask about.
Ask before changing more than one composition.
Default behavior
Describe what the skill does when the user has not added preferences of their own. Keep it to the choices you want the Assistant to make, not the mechanics of how After Effects does the work.

Approach
Understand what the user is asking for.
Inspect the relevant part of the project before changing anything.
Make the smallest useful change.
Check the result.
Boundaries
Preserve unrelated work. Do not make broad changes unless the request calls for them. When the request is unclear, make the narrowest reasonable choice or ask.

<!– Use this section for limits and exceptions: what should never be changed, what needs extra care, or when the skill should stop and ask. –>

Verify
Confirm the requested change happened and that unrelated work still behaves as before.

Report
Say what changed and mention anything that could not be done as requested.

<!– You can add, remove, or rename any section in this file. These sections are a useful default shape, not a required one. A short, clear skill beats a long, vague one. –>

​

Preflight: Check your projects for problems before render, export, sharing, handoff, etc.

---
name: preflight
description: Check an After Effects project for problems before rendering, exporting, sharing, or handing it off. Use when the user asks to preflight, check, audit, review, verify, or fix issues before delivery.
---

# Preflight

Check whether an After Effects project is ready for its intended use. Find
meaningful problems, surface important unusual states, and help resolve clear
issues instead of stopping at a report.

## Customize this skill here

> **This is where you make the skill your own.** Add the checks and standards
> that matter for your workflow below. The Assistant follows these alongside
> the default checks, and the "Default behavior" section is what happens if you
> leave this empty. When your preferences conflict with the defaults, your
> preferences win.

<!--
Add checks, preferred delivery settings, naming standards, audio expectations,
or other project rules. For rules that apply to many items, be specific about
what should count.
-->

- Deliver at 24 fps unless a comp is clearly meant to be different.
- Expect audio in any comp named with "VO" or "MIX."
- Flag any text still containing "PLACEHOLDER" or "TBD."

## Default behavior

Inspect the whole project, classify each finding as a Top issue or an Item of
note, and report only meaningful findings — never normal conditions or the
inspection process. Fix clear issues only when the user asks for fixes, and
verify every change before reporting it as fixed. When a finding depends on the
user's intent, ask rather than assume.

## Approach

1. Understand the intended delivery when it is clear.
2. Inspect the full project for checks that require project-wide coverage.
3. Classify each finding using the rules below.
4. Fix clear issues only when the user asks. Do not make silent changes.
5. Verify every change before reporting it as fixed.

When a finding depends on intent or delivery, ask the smallest useful question
rather than assuming it is wrong. Do the other checks first rather than waiting.
A required Top issue can stay in the Top issues table with a "Needs review"
status while you wait for an answer.

## What to check

**Every composition (project-wide).** Review composition-level settings across
all comps, not just the active or selected one: frame rate, duration,
dimensions and pixel aspect ratio, color depth, and other delivery-relevant
settings. For frame rate, check both whether a rate is unusual on its own and
whether it is inconsistent with other comps or the intended delivery.

**Media and dependencies (project-wide).** Look for missing or unavailable
footage, broken dependencies, broken expressions, missing or substituted fonts
when detectable, placeholder or unfinished content, and unusual source
properties. When checking source frame rates, include video, animated image
files, and image-sequence footage.

**Timeline.** For relevant comps, look for soloed layers, layers disabled or
hidden in a way that could affect the result, layers lasting only a few frames
in a much longer comp, layers starting or ending at surprising times, and other
unusual states. A very short layer may be an intentional flash or transition —
surface it for review rather than calling it wrong.

**Audio, when expected.** Check for muted or disabled audio and missing audio
that would cause expected sound not to be heard. When it is unclear whether
audio is required, ask.

Do not report normal or expected conditions: that footage is present, that
fonts are fine, that dependencies are healthy, or that layers are intentionally
disabled as part of a variant, matte, or asset-library pattern.

## How to classify findings

**Top issues** are required, high-priority findings that deserve prominent
attention:

- comps containing soloed layers
- broken expressions
- completely empty comps
- unusual or materially mismatched comp frame rates
- unusual or materially mismatched footage frame rates, including image-sequence footage
- comp settings that clearly conflict with an established project or delivery standard
- other conditions that clearly threaten the intended result

A required Top issue stays in Top issues even when it may be intentional. Mark
it "Needs review" or "May be intentional" rather than moving it to Items of
note.

**Items of note** are meaningful but lower-priority findings worth the user's
attention that do not meet a Top issue rule.

**Normal conditions are not findings.** Do not report things checked and found
normal, and do not narrate the inspection.

## Report

Report findings, not the inspection process. Keep findings concise but
informative so the report is easy to scan. Use this structure and order:

# Issues found from running the preflight

## Items of note

Include this table only when there are items of note.

| Item | Finding | Why it was noted |
|---|---|---|
| Promo Comp | 12 fps | May be intentional; review if needed |

## Top issues found

Include this table only when at least one Top issue was detected. Every
required Top issue must appear here.

| Item | Issue | Why it matters | Status |
|---|---|---|---|
| Main Comp | Broken expression | Can stop intended animation | Needs attention |
| Logo Comp | 12 fps | Unusual for this project | Needs review |

Use **Status** to distinguish clear action from uncertainty: "Needs attention,"
"Needs review," or "Fixed and verified." Do not create empty tables and do not
report normal checks. After any requested fix, update the row with the verified
status — never claim a fix without rechecking the project.

The **final line** contains only the counts, in this form, with nothing after
it:

**Top issues found: 2 · Items of note: 3**

## Boundaries and completeness

Project-wide claims require project-wide evidence. Do not say a category passed
unless the relevant comps, footage, or layers were actually covered. If the
available project information cannot establish complete coverage, say so instead
of implying uninspected items are fine.

When you cannot tell whether a finding is intentional or what the correct
setting should be, ask rather than guessing.

## Verify

Before reporting, confirm the result reflects the current project state, not an
earlier snapshot. After any change, re-inspect the affected area and confirm
the intended result before marking a finding "Fixed and verified."

## Examples

"Is this ready to render?" Run the general preflight and use the required report
structure. Keep normal conditions out.

"Check this before client delivery and fix anything obvious." Run the preflight,
fix only clearly justified issues, verify those fixes, and ask about anything
that depends on the client's intended delivery.

"This comp is 30 seconds long, but this layer is only 2 frames." Flag the short
layer for review when its purpose is unclear. Do not call it wrong merely
because it is short.

Spell Check: Check or correct text in your AE project. State your desired dialect (US, Canadian, British English, etc.) in text in the skill
 

---
name: spellcheck
description: Check or correct text in an After Effects project. Use when the user asks to spellcheck, proofread, correct, clean up, or review wording in text layers.
---

# Spellcheck

Proofread text without rewriting the user's creative voice unless asked.

## Customize this skill here

> **This is where you make the skill your own.** Add your writing preferences
> below. The Assistant follows these when checking text, and the "Default
> behavior" section is what happens if you leave this empty. When your
> preferences conflict with the defaults, your preferences win.

<!--
Replace the examples below with your own language variant, approved terms,
style, or how aggressively wording suggestions should be made.
-->

- Use American English.
- Keep the voice casual; do not formalize contractions or slang.
- Treat these as correct, not errors: our product and brand names.
- Report wording suggestions separately; do not apply them automatically.

## Default behavior

Correct clear errors when the user asked for corrections; otherwise report them
without changing the project. Always separate definite corrections from
suggestions or uncertain choices, and leave intentional creative language
alone.

## Approach

1. Identify the text the user wants checked.
2. Review it for clear errors and likely accidental wording problems.
3. Correct clear errors when correction was requested; otherwise report them without changing the project.
4. Separate definite corrections from suggestions or uncertain choices.

## What to catch and what to leave alone

Catch clear, likely-accidental problems:

- spelling mistakes
- grammar and punctuation errors
- obvious typos
- inconsistent capitalization
- awkward or unclear wording when it is clearly accidental

Leave intentional language alone. Do not "correct" names, brands, product
names, places, slang, invented words, deliberate stylization, or unusual
capitalization when they appear intentional. Do not change wording simply
because another style would be possible. When a correction is uncertain, flag
it rather than silently changing it.

## Boundaries

Change only the text the user asked about. When correction was not requested,
report findings without modifying the project.

## Verify

After corrections, check the affected text again and confirm the intended
wording is still present.

## Report

List clear corrections separately from suggestions. If you changed the project,
say what changed; if you only reviewed, report findings without implying any
change was made.

## Examples

"Spellcheck this." Check the relevant text and report clear errors without
changing the project unless correction was requested.

"Fix the copy in these titles." Correct clear errors in the specified text
layers and report what changed.

"Proofread this, but keep the voice casual." Correct clear errors while
preserving the user's style and intentional informal language.

Hello friends! I have returned and have a whopper of an update for you with what we’ve added since late August! Get ready.

First -- user-provided skills are available in beta build 27.1 x8 (so, today’s build!) and onward.

Skills are reusable instructions that tell the Assistant how you want a particular task handled. You can use them to define your project organization standards, your approach to troubleshooting, your preferences for animation work, or creating reusable instructions for tasks the Assistant already supports. You don’t need coding experience! Write your preferences and boundaries in plain language.

To get started, open the three-dot menu in the Assistant panel, choose Skills, and click “Open Skills folder.” 

To create your own skill, use the template below. Give the folder and the “name” field at the top of SKILL.md the same name. A skill named “my-skill” is called by typing /my-skill in the Assistant.Keep each skill focused on one job. A project-organizing skill should describe how you want projects organized; animation and troubleshooting preferences belong in separate skills. A skill’s instructions apply to its task, not to every interaction with the Assistant.

Skills Readme:

​
# After Effects skills

Starter skills for the After Effects Assistant. A skill is a small set of instructions that shapes how the Assistant works when you call it with `/skill-name`. It applies only to that skill's task — other skills are unaffected.

You don't need coding experience to use these files. Each skill includes default settings you can use without changes, plus a clearly marked place to make it your own. Open a `SKILL.md` and tell the Assistant your preferences, standards, and boundaries in plain language.

## How a skill is organized

Each skill lives in its own folder with one required file, `SKILL.md`:

```text
skills/
|-- preflight/
|   `-- SKILL.md
|-- spellcheck/
|   `-- SKILL.md
`-- my-skill/
    `-- SKILL.md
```

The folder name, the `name` in the file settings at the beginning of `SKILL.md`, and the `/skill-name` command all match. For a folder named `my-skill`:

```yaml
---
name: my-skill
description: Improve something in a clear, reusable way.
---
```

- **`name`** is the command you type after `/`.
- **`description`** helps the Assistant recognize when a request belongs to the skill. Write it the way you'd naturally ask for the task, rather than in technical terms.

## Included skills

- `/preflight` — check a project before rendering or delivery
- `/spellcheck` — proofread and correct text in After Effects

## Customize a skill

Every skill has two layers:

- **Default behavior** — what the skill does before you customize it.
- **Your preferences** — the rules you add to fit your workflow. When the two conflict, your preferences win.

To customize, open a skill's `SKILL.md` and find the section marked **"Customize this skill here."** Add short, specific rules, for example:

- "Use Canadian English."
- "Prefer subtle overshoot."
- "Never delete original layers automatically."
- "Keep our approved layer prefixes."

Each skill includes a few example preferences to show what works well — replace them with your own. You don't need to rewrite the whole file; start by editing the examples.

Skills are written in plain language on purpose: they describe the result you want and the choices the Assistant should make, not the mechanics of operating After Effects. The Assistant already knows how to use After Effects, so "always leave a 24 fps master" or "never rename my control layers" belongs in a skill; step-by-step panel instructions do not.

When you're ready for more, you can edit beyond the preferences. The sections used in these skills — What it does, Customize, Default behavior, Approach, Boundaries, Verify, Report — are a starting point, not a rule. Add, remove, or reorder sections to fit. A short, clear skill works better than a long, vague one.

## Create a skill

Start from `SKILL-TEMPLATE.md`, or copy an existing skill folder:

1. Create a folder for the skill.
2. Add `SKILL.md` inside it.
3. Set `name` in the file settings at the beginning of `SKILL.md` to match the folder name. This becomes the `/` command.
4. Write a `description` covering what the skill does and the requests that should trigger it.
5. Add your preferences in the Customize section.
6. Keep the defaults, boundaries, and examples that fit; remove the rest.

## Keep skills focused

A skill works best with one clear job. A project-organizing skill should describe how you want projects organized, not also try to be a general animation or troubleshooting skill. Keep unrelated tasks in their own skills.

Skill Template:

<!--
HOW TO USE THIS TEMPLATE

A skill lives in its own folder, and the instruction file must be named SKILL.md.

    my-skill/
    `-- SKILL.md

Keep three things the same:
  - the folder name           (my-skill)
  - the frontmatter "name"    (name: my-skill)
  - the slash command         (/my-skill)

This template is written as a small, complete example skill you can copy and
change. Replace the parts that describe the example with your own task.
-->

---
name: my-skill
description: Describe what this skill does and the kinds of requests that should use it. Write it the way a user would ask for the task, not in terms of After Effects internals.
---

# My Skill

One sentence describing the result this skill should achieve.

## Customize this skill here

> **This is where you make the skill your own.** Add your preferences, rules,
> standards, or exceptions below. The Assistant follows these when the skill
> runs, and everything under "Default behavior" is what happens if you leave
> this empty. When your preferences conflict with the defaults, your
> preferences win.

<!--
Good preferences are short and specific. You can add a preference, a rule, a
standard, or an exception. You do not need to rewrite the rest of the skill.

Replace the examples below with your own.
-->

- Prefer subtle results over dramatic ones.
- Never change anything I did not ask about.
- Ask before changing more than one composition.

## Default behavior

Describe what the skill does when the user has not added preferences of their
own. Keep it to the choices you want the Assistant to make, not the mechanics
of how After Effects does the work.

## Approach

1. Understand what the user is asking for.
2. Inspect the relevant part of the project before changing anything.
3. Make the smallest useful change.
4. Check the result.

## Boundaries

Preserve unrelated work. Do not make broad changes unless the request calls
for them. When the request is unclear, make the narrowest reasonable choice or
ask.

<!--
Use this section for limits and exceptions: what should never be changed, what
needs extra care, or when the skill should stop and ask.
-->

## Verify

Confirm the requested change happened and that unrelated work still behaves as
before.

## Report

Say what changed and mention anything that could not be done as requested.

<!--
You can add, remove, or rename any section in this file. These sections are a
useful default shape, not a required one. A short, clear skill beats a long,
vague one.
-->

Other Assistant updates since August

• Speed: we turned on the afterburners! You should now see a considerable decrease in the time it takes for the Assistant to complete a response to your prompt. Hit thumbs up/down on your response and give us feedback on how the new speed feels.

• Composition management: the Assistant can now open compositions and is instructed to open newly created compositions automatically instead of leaving them unopened in the Project panel.

• Text editing: the Assistant can edit a specific range within a text layer rather than replacing all of its text.

• Motion Graphics templates and Essential Graphics: expanded support for inspecting Essential Graphics settings, exposing properties and media, renaming controls, exporting MOGRTs, and working with Essential Properties.

• Shape layers: expanded support for adding, deleting, renaming, reordering, and enabling shape operators, plus editing stroke dashes.

• Expressions and animation: renaming layers, project items, or effects through the Assistant now updates expressions that reference them. We’ve also expanded keyframe interpolation support and improved checkbox expressions.

• Project organization: improved support for organizing project items, including moving items back to the project root.

• Render output: when you don’t specify an output path, the Assistant uses a default render location. On Windows: save your project first, or the Assistant won’t be able to set that default render location. If you haven’t saved your project on Mac, the default render location will be the desktop.

We’ve also updated the Assistant’s guidance to use reasonable defaults, resolve references from the current project, ask fewer unnecessary questions, and report the concrete changes it made in your project.

Please continue providing feedback! Tell us which instructions worked, where the Assistant didn’t follow them, and what you expected instead -- best path for reporting specific workflow issues is to thumbs up or down the Assistant’s prompt response and provide comments on your experience. That feedback will go straight to us and we can action it ASAP, even if you won’t hear back from us directly.

Hi all! Quick check in before I head out of office for a few weeks. I wanted to give an update on some bug fixes, speed improvements, and new capabilities since my last post.
 

Bug Fixes:

• The Assistant no longer scans or enumerates files outside your project to resolve a path — it stays scoped to the task at hand. Big thanks to ​@Shebbe  (and ​@Roland Kahlenberg ) for bringing the issue to our attention.
• Tools no longer drop out mid-session
• Fixed the Assistant failing to connect if the panel was open at startup
• Panel loads reliably, and is now fully localized (AKA you can use any language you want when working with the Assistant)

New capabilities:

• Text manipulation — edit and drive text layers directly through the Assistant
• Open Comp — the Assistant can open the comp it's working in
• Generates and imports files directly instead of just linking you to help docs

Better results on real work:

• Renaming a layer no longer breaks expressions that reference it — the Assistant auto-fixes the references
• Correct 3D renderer is now chosen for parametric/mesh scenes
• Better handling of shape-layer hierarchies — selection awareness, nested groups, and driving the right content property (e.g. Rectangle Path → Size instead of Scale)
• Checkbox-driven expressions now include the required .value
• Fewer incorrect "that's out of scope" responses