Agent-skill-layerproof slides

Public API slide editing (X-API-KEY). Edit slide images with AI, accept/revert edits, object removal, text extraction, save Konva nodes. Types follow PublicApiSlideController (/api/v2/projects/{projectId}/slides).

install

source · Clone the upstream repo

git clone https://github.com/compilet-dev/agent-skill-layerproof

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/compilet-dev/agent-skill-layerproof "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/slides" ~/.claude/skills/compilet-dev-agent-skill-layerproof-slides && rm -rf "$T"

manifest: skills/slides/SKILL.md

source content

Skill: Slide Editing

Description

Modify slides after generation. This skill documents the public API at

/api/v2/projects/{projectId}/slides

(PublicApiSlideController). All slide endpoints require

projectId

and

slideId

in the path. Authenticate with

X-API-KEY

header.

TypeScript types (request / response)

Mirrors

PublicApiSlideController

(/api/v2/projects/{projectId}/slides) data classes.

// Shared: cropped region for region-based editing. Must lie fully within image bounds.
// Constraints: x >= 0, y >= 0, width > 0, height > 0; x + width <= image width, y + height <= image height.
type CroppedRegion = {
  x: number;
  y: number;
  width: number;
  height: number;
};

// --- Edit Slide Image (POST) — async, poll jobs ---
type SlideImageEditRequest = {
  /** AI instruction for the edit (required, non-blank) */
  instruction: string;
  /** Path to input image in project working dir (required, non-blank) */
  input_image_path: string;
  /** Optional extra reference image paths */
  other_reference_image_paths?: string[];
  /** Optional region to edit; only this region is modified */
  cropped_region?: CroppedRegion;
};

// --- Accept Image Edit (POST) ---
type AcceptImageEditRequest = {
  live_object_id: string;   // UUID from image-edit or object-removal response (required)
  target_node_id?: string;
  override_image_path?: string;  // Optional path to use instead of workflow output
};

// --- Revert Slide (POST) ---
type RevertSlideRequest = {
  history_entry_id: string;  // UUID from slide history (required)
  node_id?: string;
};

// --- Object Removal (POST) — async, poll jobs ---
type ObjectRemovalRequest = {
  input_image_path: string;  // required, non-blank
  mask_path?: string;
  cropped_region?: CroppedRegion;
};

// --- Extract Text (POST) — async, poll jobs ---
type ExtractTextRequest = {
  node_id?: string;
  node_image_path?: string;
};

// --- Save Konva Nodes (PUT) ---
type SaveKonvaNodesRequest = {
  konva_nodes: Record<string, unknown>;  // required
  konva_order: string[];                  // required
  base_snapshot_id?: string;  // UUID, optional
};

// --- Responses ---
type TriggerWorkflowResponse = {
  activity_id: string;   // UUID – poll GET /api/v2/jobs/{activityId}
  workflow_type: string;
  live_object_id: string; // UUID – pass to accept-image-edit when done
};

type AcceptImageEditResponse = {
  id: string;             // slide UUID
  slide_section_id: string;
  image_path: string | null;
  status: string;
};

Edit Slide Image (async)

Request body:

SlideImageEditRequest

. Response:

TriggerWorkflowResponse

Triggers AI-powered slide image editing. Upload input image first, then call with

instruction

and

input_image_path

. Poll

GET /api/v2/jobs/{activityId}

for status; when DONE, call accept-image-edit with

live_object_id

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/image-edit" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"instruction":"Make the background blue","input_image_path":"/slides/slide-1.png"}'

With optional region and reference images:

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/image-edit" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"instruction":"Replace this area","input_image_path":"/slides/slide-1.png","cropped_region":{"x":0,"y":0,"width":200,"height":100},"other_reference_image_paths":["/ref.png"]}'

Accept Image Edit

Request body:

AcceptImageEditRequest

. Response:

AcceptImageEditResponse

Call after the image-edit or object-removal workflow completes successfully. Use

live_object_id

from the trigger response.

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/accept-image-edit" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"live_object_id":"<live_object_uuid>"}'

Revert Slide

Request body:

RevertSlideRequest

. Response:

AcceptImageEditResponse

Reverts the slide to a previous version using

history_entry_id

from slide history.

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/revert" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"history_entry_id":"<history_entry_uuid>"}'

Remove Objects (async)

Request body:

ObjectRemovalRequest

. Response:

TriggerWorkflowResponse

Triggers AI-powered object removal. Optional

mask_path

for mask-based removal. Poll jobs; when DONE, call accept-image-edit with

live_object_id

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/object-removal" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"input_image_path":"/slides/slide-1.png"}'

With mask:

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/object-removal" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"input_image_path":"/slides/slide-1.png","mask_path":"/masks/mask.png"}'

Extract Text (async)

Request body:

ExtractTextRequest

(optional). Response:

TriggerWorkflowResponse

Performs OCR, removes text from image via inpainting, creates editable Konva text nodes. Poll

GET /api/v2/jobs/{activityId}

for status.

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/extract-text" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{}'

With optional node:

curl -X POST "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/extract-text" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"node_id":"node-1","node_image_path":"/nodes/node1.png"}'

Save Konva Nodes

Request body:

SaveKonvaNodesRequest

. Response: 200, no body.

Persists Konva canvas nodes and order (positions, layer order, flattened image path).

curl -X PUT "$LAYERPROOF_BASE_URL/api/v2/projects/<project_id>/slides/<slide_id>/konva-nodes" \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: $LAYERPROOF_API_KEY" \
  -d '{"konva_nodes":{},"konva_order":[]}'

Agent behavior

When the user asks to edit slides (image edit, accept edit, revert, remove objects, extract text, save canvas), do the following.

1. Choose the right endpoint

User intent	Endpoint	Method
Edit slide image with AI instruction	`/api/v2/projects/{projectId}/slides/{slideId}/image-edit`	POST
Accept edited image after workflow completes	`/api/v2/projects/{projectId}/slides/{slideId}/accept-image-edit`	POST
Revert slide to previous version	`/api/v2/projects/{projectId}/slides/{slideId}/revert`	POST
Remove objects from slide image (AI)	`/api/v2/projects/{projectId}/slides/{slideId}/object-removal`	POST
Extract text from slide (OCR + inpainting)	`/api/v2/projects/{projectId}/slides/{slideId}/extract-text`	POST
Save Konva canvas nodes/order	`/api/v2/projects/{projectId}/slides/{slideId}/konva-nodes`	PUT

All paths require

projectId

and

slideId

(replace placeholders with actual UUIDs).

2. Build and run the request

Step 1 — Check environment variables first. Before running any curl command, verify both

LAYERPROOF_BASE_URL

and

LAYERPROOF_API_KEY

are set on the user's machine:

if [[ -z "${LAYERPROOF_BASE_URL}" ]]; then
  echo "ERROR: LAYERPROOF_BASE_URL is not set."
  return 1
fi
if [[ -z "${LAYERPROOF_API_KEY}" ]]; then
  echo "ERROR: LAYERPROOF_API_KEY is not set."
  return 1
fi

If running from a project directory with a

.env.local

file, load it first:

if [[ -f .env.local ]]; then
  set -a
  source .env.local
  set +a
fi

Step 2 — Auth: Every request must include

X-API-KEY: $LAYERPROOF_API_KEY

. Read

LAYERPROOF_BASE_URL

and

LAYERPROOF_API_KEY

from the environment; if missing, tell the user to set them. Step 3 — Path: Resolve

project_id

and

slide_id

from context (e.g. from project/slide-deck list or user input). If missing, ask the user. Step 4 — POST/PUT: Build JSON body from the types above. Use

-X POST

-X PUT

-H "Content-Type: application/json"

, and

-d '...'

. Run the curl and show the result.

3. After async endpoints (image-edit, object-removal, extract-text)

Response includes
```
activity_id
```
and
```
live_object_id
```
. Tell the user the workflow was started.

Suggest polling

GET $LAYERPROOF_BASE_URL/api/v2/jobs/<activity_id>

until

status

DONE

CANCELED

For image-edit and object-removal: when DONE, tell the user to call accept-image-edit with
```
live_object_id
```
to apply the result.

4. Response handling

Always show the raw JSON response in a JSON code block; do not convert to a table.
If the response contains a URL for an image (e.g.
```
image_path
```
), show the image and the JSON.
On error (4xx/5xx), show the response body and status code; suggest fixing API key, projectId/slideId, or request body.

5. Example workflows

Workflow A — User: "Edit slide image in project X, slide Y: make the background darker."

Choose

POST /api/v2/projects/{projectId}/slides/{slideId}/image-edit

Resolve projectId and slideId (from user or ask).

Build body:

{"instruction":"Make the background darker","input_image_path":"/path/to/slide/image"}

(user may need to provide image path).

Run curl; show JSON. Mention polling jobs with
```
activity_id
```
and calling accept-image-edit with
```
live_object_id
```
when done.

Workflow B — User: "Edit two slides: darker background on slide 1, remove the logo from slide 3; then revert slide 1 if I don’t like it."

Resolve projectId; get slide ids from slide-deck GET deck (e.g.
```
slides[0].id
```
,
```
slides[2].id
```
).

For slide 1: POST

.../slides/{slideId1}/image-edit

with

{"instruction":"Make the background darker"}

; capture

activity_id_1

and

live_object_id_1

. For slide 3: POST

.../slides/{slideId3}/object-removal

with region/instruction; capture

activity_id_3

and

live_object_id_3

Poll
```
GET /api/v2/jobs/{activity_id_1}
```
and
```
.../jobs/{activity_id_3}
```
until both DONE. If either fails, report which slide and reason.

POST

.../slides/{slideId1}/accept-image-edit

with

{"live_object_id":"<live_object_id_1>"}

; POST

.../slides/{slideId3}/accept-image-edit

with

{"live_object_id":"<live_object_id_3>"}

. Show updated slide images (GET deck or slide detail).

If user says "revert slide 1": GET slide history (if available) for slideId1 to get
```
history_entry_id
```
; POST
```
.../slides/{slideId1}/revert
```
with
```
{"history_entry_id":"<id>"}
```
. Confirm with GET deck.

Workflow C — User: "Extract text from slide 2 and replace the old text (OCR + inpainting)."

Resolve projectId and slideId for slide 2. POST
```
.../slides/{slideId}/extract-text
```
(body per API); capture
```
activity_id
```
.
Poll
```
GET /api/v2/jobs/{activity_id}
```
until DONE. When DONE,
```
output
```
may contain extracted text or updated asset reference.
If a follow-up accept or apply step is required (e.g. accept-image-edit), use the returned
```
live_object_id
```
; otherwise show the result (e.g. updated slide or transcript) from GET deck.

Response format (required)

(if response contains url to show image) please show image and show json response instead of table
Always show the raw JSON response (verbatim) in a JSON code block.
If the response contains a URL for an image, render/show the image and also show the JSON response (do not convert to a table).