screen-memory
/screen-memory
Screen Memory — the disabled-by-default, local-only desktop buffer of recent screen/app/window context, plus its status and query actions. Use when the user asks what was on screen recently, or when reading, exporting, enabling, or describing Screen Memory.
- Author
- builderio ↗
- Repository
- agent-native ↗
- Licence
- MIT
- Last repository change
- 27.10.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Screen Memory — the disabled-by-default, local-only desktop buffer of recent screen/app/window context, plus its status and query actions. Use when the user asks what was on screen recently, or when reading, exporting, enabling, or describing Screen Memory.
Original author description
Screen Memory — the disabled-by-default, local-only desktop buffer of recent screen/app/window context, plus its status and query actions. Use when the user asks what was on screen recently, or when reading, exporting, enabling, or describing Screen Memory.
How it works
- Rule
- Reading it from the in-app agent
- User control and external agents
- Earlier screen time (lab)
- Related skills
SKILL.md preview
---
name: screen-memory
description: >-
Screen Memory — the disabled-by-default, local-only desktop buffer of recent
screen/app/window context, plus its status and query actions.
Use when the user asks what was on screen recently, or when reading,
exporting, enabling, or describing Screen Memory.
---
# Screen Memory
## Rule
Screen Memory is a disabled-by-default, local-only desktop buffer of recent
screen, app, and window context. Raw segments are never uploaded or shared, and
never describe Screen Memory as hosted, shared, exhaustive, or enabled by
default.
The one exception is an owner's explicit request for earlier screen time (see
below). Then a chosen window of Rewind history is uploaded as an owner-only
recording attached to one Clip. That Clip follows its normal share rules, and
sharing the Clip shares its earlier screen time with it. The rest of the buffer
stays local.
## Reading it from the in-app agent
1. Call `get-screen-memory-status` before relying on it at all.
2. Then call `query-screen-memory-context` for bounded recent snippets when
local context files are present.
If the local Screen Memory MCP built-in is connected, the agent may also use
`screen_memory_status`, `screen_memory_recent_context`, and
`screen_memory_recent_segments`. Only inspect or export segment file paths when
the user explicitly asks.
## User control and external agents
Users enable, pause, export, and clear the buffer from the desktop tray
settings. External local agents can read recent app/window context through
`agent-native mcp screen-memory`.
While Rewind is off or user-paused, unpinned history past the time limit is
deleted at launch and within 10 minutes while the app runs. Pausing keeps
history inside the limit. Active capture prunes on segment rotation. Active
or suspended backend-owned segments stay protected until finalized. Separately,
finalized segments pinned by Clip export or meeting-history work stay protected
from time-limit and byte-cap pruning until all consumer pins are released.
Cleanup includes history from older versions in this bundle's data directory.
Other bundle identifiers have separate stores and are not swept. Saved Clips
and agent handoffs are never part of this.
Do not upload raw Screen Memory segments or treat them as shareable Clips unless
the user explicitly exports and imports them.
## Earlier screen time (lab)
Behind the `clips.lookback-context` lab (`CLIPS_LOOKBACK_CONTEXT`, on by
default). This only makes the option available: Rewind is off by default, and
the per-device lookback duration starts at `0` / Off. The user must enable
Rewind and choose a duration before earlier screen time is captured. When the
lab is off, do not offer this feature.
This is not Screen Memory. It is passive metadata on one clip: screen history
from before that recording started, attached to the clip as its own context.
The window is never stitched into the clip's video, and the clip's video is
unchanged. The older editor option that stitched Rewind history into the video
("Add what happened before") is removed. Do not offer it.
An item has two windows, both inside the 5 minutes before the recording
started:
- The original window (`originalStartedAt` / `originalEndedAt`) is the widest
allowed, the last 5 minutes. It is fixed when the item is requested.
- The current window (`startedAt` / `endedAt`) is the part the owner selected.
It starts as the last `requestedSeconds` (1 to 300) before the recording
started. Any window inside the original, at least 1 second long, is allowed.
It follows its clip's normal share rules and nothing is special. Requesting it
does not change the clip's visibility. If the clip is public, shared with the
organization, or shared with people, the same viewers see the earlier screen
time: sharing the clip shares both as one bundle, and the clip's password and
expiry apply to it too. Only the owner can remove it. The request's `endedAt` may be up to 10 minutes after the Clip row
is created because native setup and the countdown happen after row creation. It
may be up to 120 seconds before row creation for client clock skew. Timestamps
outside those bounds fail with `recording_context_invalid_window`.
The footage is a separate recording owned by the clip owner, uploaded from this
device through the Rewind handoff. It stays owner-only by itself; viewers are
served it through the clip they can access, never by opening it directly. Its source app name is `Clips Rewind`; the footage check
refuses any other recording. The export handshake goes through
`update-recording-context`: `processing` claims the item and may name the
footage recording it created (`mediaRecordingId`); `ready` must name that same
recording, or it fails with `recording_context_footage_mismatch`; `failed`
releases the claim's footage. Changing the window returns the item to `pending`
and releases any reservation, and the desktop re-exports and uploads only the
new window. Removing the item trashes that footage recording and any footage an
export is still reserving.
The desktop editor builds a local preview of the original window with the
`rewind_preview_window` Tauri command, and deletes it with
`rewind_preview_discard`. The preview is never uploaded, and no action exposes
it.
On the web, the clip's Transcript tab shows the context above the transcript
when the lab is on and the clip has an item. Editing happens in the desktop app.
Read it with:
- `list-recording-context` (viewer access) returns the clip's active item: its
window, `status` (`pending`, `processing`, `ready`, or `failed`), and any
error. Removed items are excluded. Only a `ready` item has footage. A
`pending` item captured on another device waits until that device saves it,
because only that device can export its footage.
- `list-pending-recording-context` is for the desktop export worker, not for
answering questions. It lists pending requests on clips the signed-in user
owns. Pass `excludeIds[]` for items this device cannot process, so they do
not hold up the batch.
Writes are owner-only: `request-recording-context`,
`update-recording-context`, `set-recording-context-window`, and
`remove-recording-context`. The desktop worker drives exports and status
changes. Change the window or remove the context only when the owner asks.
## Related skills
- `recording` — the hosted capture path Screen Memory is not part of.
- `security` — why local-only context stays local.
- `context-awareness` — the supported way the agent learns what is on screen
inside Clips.
Versions
The database has no other skills named screen-memory.
Database snapshot: 13.11.2026 · package version 4e31868eae2d (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90memory
Use when the user references prior work, asks what happened recently, needs context about the project or person, or when you should store notes from a call, meeting, or important conversation. Also activate when the user says "remember this" or asks you to search memory. - 95macos-control-skill
Triggered when the user asks the AI to directly operate their macOS desktop: open an app, click a button, fill a form inside a native window, read what's currently on screen, automate a local GUI flow. Typical phrasing: "in Finder, please…", "click X in the Y app", "what's on my… - 98openai-screenshot
Use when the user explicitly asks for a desktop or system screenshot (full screen, specific app or window, or a pixel region), or when tool-specific capture capabilities are unavailable and an OS-level capture is needed. Originally from OpenAI's curated skills catalog.