Generate the Copilot and Cline skill paths from one source

.agents/skills/README.md

@@ -0,0 +1,14 @@
+# Generated skills
+
+Do not edit these files. They are copied byte for byte from
+`plugins/dreamlayer/skills/`, the canonical source, so that Cursor, OpenCode, GitHub Copilot and the npx skills installer load the same
+workflows as every other host.
+
+Change a workflow in the canonical source, then regenerate:
+
+```sh
+python3 scripts/sync_skills.py
+```
+
+`python3 scripts/sync_skills.py --check` fails when any copy has drifted, and
+`python3 scripts/validate.py` runs that check.

.cline/skills/README.md

@@ -0,0 +1,14 @@
+# Generated skills
+
+Do not edit these files. They are copied byte for byte from
+`plugins/dreamlayer/skills/`, the canonical source, so that Cline load the same
+workflows as every other host.
+
+Change a workflow in the canonical source, then regenerate:
+
+```sh
+python3 scripts/sync_skills.py
+```
+
+`python3 scripts/sync_skills.py --check` fails when any copy has drifted, and
+`python3 scripts/validate.py` runs that check.

.cline/skills/batch-photo-editing/SKILL.md

@@ -0,0 +1,11 @@
+---
+name: batch-photo-editing
+description: "Apply a specified edit, background removal or upscaling task to a list of local images with DreamLayer. Use for a bounded batch with per-file results and a total credit budget."
+---
+
+# batch photo editing
+
+Read [the workflow contract](references/workflow-contract.md) for setup, billing, recovery and output checks.
+
+Enumerate input files and assign unique new outputs before uploading. Estimate total operations from current capabilities; an edit plus cutout is two paid operations per image. Maintain a per-item ledger of input, asset ID, saved request, idempotency key, execution ID, status and output. Process sequentially initially; stop new submissions on insufficient credits or exhausted budget. Resume existing executions after interruption. Skip already completed outputs rather than changing paths and charging again. Inspect a sample first and all outputs for tasks where fidelity matters. The batch is orchestration over individual operations, not a separate DreamLayer batch endpoint.
+

.cline/skills/batch-photo-editing/references/workflow-contract.md

@@ -0,0 +1,46 @@
+# DreamLayer workflow contract
+
+Use the installed DreamLayer MCP tools, or the pinned CLI when MCP is unavailable.
+Read the actual tool schema before calling. Never pass fields a client does not expose.
+
+## Setup
+Requires Node.js >=22.12 and a DreamLayer API key from https://platform.dreamlayer.io.
+Install the supported sprite-capable beta: `npm install -g dreamlayer@0.4.0-beta.3`.
+MCP: `npx -y @dreamlayer/mcp@0.4.0-beta.4`. Supply DREAMLAYER_API_KEY to the server process using the host's secret settings. Never commit keys or put them in prompts.
+The stable npm tag remains 0.3.0. Do not claim it supports this complete workflow.
+
+## Before paid work
+1. Read dreamlayer_capabilities and dreamlayer_balance (or CLI capabilities/balance --json). If either fails, resolve it before spending.
+2. Confirm the requested operation is advertised. Ordinary completed image operations currently cost one credit each; obtain the live cost from capabilities. A multi-step edit and cutout costs two operations.
+3. Use the user's already-authorized operation, number of outputs and credit budget. Ask only if missing. Never buy credits or expand the batch implicitly.
+4. For file operations, verify input readability and a new output path before submitting. Upload once using dreamlayer_upload_image; preserve its input_asset_id.
+5. Save a unique idempotency_key and the complete request locally before each generation. Do not store credentials in the job record.
+
+## Execute and recover
+Select operation explicitly: text_to_image; image_to_image; background_remove; upscale; sprite_sheet.
+All except text_to_image require input_asset_id. Generate returns execution_id, status, asset, question and last_event_id.
+Preserve IDs/cursor after each response. For running work, call dreamlayer_execution or dreamlayer_events; do not generate a replacement after a timeout.
+A needs_input result is a question, not a failed generation. Continue using conversation_id and respond.
+For completed work, dreamlayer_download writes an absolute new local path. Never treat a returned asset URL as proof that the local file exists.
+On a local download error, repair the destination and download the same execution.
+File-based CLI retries upload a new asset; use status and download after uncertainty rather than rerunning the paid command.
+Stop at insufficient credits, invalid request, unsupported operation, or an unresolved contract error. Keep recoverable state; do not loop paid retries.
+
+## Output acceptance
+Inspect the saved image or preview. Check dimensions, transparency if requested, text spelling, reference fidelity and the intended use. Label uninspected outputs.
+PNG output does not establish SVG/vector structure, pixel-perfect art, trademark clearance, print compliance, exact product fidelity or consistent animation.
+API documentation: https://docs.dreamlayer.io/agent-api
+Recovery: https://docs.dreamlayer.io/agent-api/jobs-and-events
+
+## CLI examples
+```sh
+dreamlayer capabilities --json
+dreamlayer balance --json
+dreamlayer generate "Original image brief" --aspect 1:1 --idempotency-key UNIQUE_SAVED_KEY --out concept.png --json
+dreamlayer edit reference.png "Change the background to a simple studio scene" --idempotency-key UNIQUE_SAVED_KEY --out edited.png --json
+dreamlayer cutout reference.png --idempotency-key UNIQUE_SAVED_KEY --out cutout.png --json
+dreamlayer status EXECUTION_ID --json
+dreamlayer download EXECUTION_ID --out recovered.png --json
+```
+Placeholder IDs and keys must be replaced. Check CLI --help for supported flags; --max-credits is for sprite jobs.
+

.cline/skills/generate-or-edit-image/SKILL.md

@@ -0,0 +1,11 @@
+---
+name: generate-or-edit-image
+description: "Generate an original image, edit a local reference, remove a background, or upscale an image with DreamLayer. Use for visual assets in an app, website or creative project when the user wants image files."
+---
+
+# generate or edit image
+
+Read [the workflow contract](references/workflow-contract.md) for setup, billing, recovery and output checks.
+
+Establish the visual brief, output count, aspect ratio, destination and authorized credits. Select the operation by the requested change; an edit needs a reference, a new image does not. Use background_remove for transparency rather than assuming a prompt creates alpha. Upscale doubles both dimensions; input longest side must be <=2048 and output is capped at 4096. Supported aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4. Save files into the project, inspect, and report dimensions and remaining limitations. For logos, product photos, marketing and print artwork, read the matching installed task skill if present.
+

.cline/skills/generate-or-edit-image/references/workflow-contract.md

@@ -0,0 +1,46 @@
+# DreamLayer workflow contract
+
+Use the installed DreamLayer MCP tools, or the pinned CLI when MCP is unavailable.
+Read the actual tool schema before calling. Never pass fields a client does not expose.
+
+## Setup
+Requires Node.js >=22.12 and a DreamLayer API key from https://platform.dreamlayer.io.
+Install the supported sprite-capable beta: `npm install -g dreamlayer@0.4.0-beta.3`.
+MCP: `npx -y @dreamlayer/mcp@0.4.0-beta.4`. Supply DREAMLAYER_API_KEY to the server process using the host's secret settings. Never commit keys or put them in prompts.
+The stable npm tag remains 0.3.0. Do not claim it supports this complete workflow.
+
+## Before paid work
+1. Read dreamlayer_capabilities and dreamlayer_balance (or CLI capabilities/balance --json). If either fails, resolve it before spending.
+2. Confirm the requested operation is advertised. Ordinary completed image operations currently cost one credit each; obtain the live cost from capabilities. A multi-step edit and cutout costs two operations.
+3. Use the user's already-authorized operation, number of outputs and credit budget. Ask only if missing. Never buy credits or expand the batch implicitly.
+4. For file operations, verify input readability and a new output path before submitting. Upload once using dreamlayer_upload_image; preserve its input_asset_id.
+5. Save a unique idempotency_key and the complete request locally before each generation. Do not store credentials in the job record.
+
+## Execute and recover
+Select operation explicitly: text_to_image; image_to_image; background_remove; upscale; sprite_sheet.
+All except text_to_image require input_asset_id. Generate returns execution_id, status, asset, question and last_event_id.
+Preserve IDs/cursor after each response. For running work, call dreamlayer_execution or dreamlayer_events; do not generate a replacement after a timeout.
+A needs_input result is a question, not a failed generation. Continue using conversation_id and respond.
+For completed work, dreamlayer_download writes an absolute new local path. Never treat a returned asset URL as proof that the local file exists.
+On a local download error, repair the destination and download the same execution.
+File-based CLI retries upload a new asset; use status and download after uncertainty rather than rerunning the paid command.
+Stop at insufficient credits, invalid request, unsupported operation, or an unresolved contract error. Keep recoverable state; do not loop paid retries.
+
+## Output acceptance
+Inspect the saved image or preview. Check dimensions, transparency if requested, text spelling, reference fidelity and the intended use. Label uninspected outputs.
+PNG output does not establish SVG/vector structure, pixel-perfect art, trademark clearance, print compliance, exact product fidelity or consistent animation.
+API documentation: https://docs.dreamlayer.io/agent-api
+Recovery: https://docs.dreamlayer.io/agent-api/jobs-and-events
+
+## CLI examples
+```sh
+dreamlayer capabilities --json
+dreamlayer balance --json
+dreamlayer generate "Original image brief" --aspect 1:1 --idempotency-key UNIQUE_SAVED_KEY --out concept.png --json
+dreamlayer edit reference.png "Change the background to a simple studio scene" --idempotency-key UNIQUE_SAVED_KEY --out edited.png --json
+dreamlayer cutout reference.png --idempotency-key UNIQUE_SAVED_KEY --out cutout.png --json
+dreamlayer status EXECUTION_ID --json
+dreamlayer download EXECUTION_ID --out recovered.png --json
+```
+Placeholder IDs and keys must be replaced. Check CLI --help for supported flags; --max-credits is for sprite jobs.
+

.cline/skills/godot-sprite-import/SKILL.md

@@ -0,0 +1,11 @@
+---
+name: godot-sprite-import
+description: "Import an existing DreamLayer sprite ZIP into a Godot 4 demo and play the animation. Use when the user already has generated frames and wants a runnable game-engine example."
+---
+
+# godot sprite import
+
+Read [the workflow contract](references/workflow-contract.md) for setup, billing, recovery and output checks.
+
+This is local import and spends no credits. Use scripts/import_sprite.py with --engine godot --zip /absolute/walk.zip --out /absolute/new-demo. It reads PNG frames and atlas durations from the bundle, validates them and creates a Godot 4 project with AnimatedSprite2D. Run godot --headless --path /absolute/new-demo --editor --quit, then open the project and inspect playback. Preserve returned frame durations; do not assume constant FPS. Use the generated demo as a reference when importing into an existing project, keeping existing settings intact. If Godot is unavailable, mark engine validation pending; successful file generation alone does not prove engine playback.
+

.cline/skills/godot-sprite-import/references/workflow-contract.md

@@ -0,0 +1,46 @@
+# DreamLayer workflow contract
+
+Use the installed DreamLayer MCP tools, or the pinned CLI when MCP is unavailable.
+Read the actual tool schema before calling. Never pass fields a client does not expose.
+
+## Setup
+Requires Node.js >=22.12 and a DreamLayer API key from https://platform.dreamlayer.io.
+Install the supported sprite-capable beta: `npm install -g dreamlayer@0.4.0-beta.3`.
+MCP: `npx -y @dreamlayer/mcp@0.4.0-beta.4`. Supply DREAMLAYER_API_KEY to the server process using the host's secret settings. Never commit keys or put them in prompts.
+The stable npm tag remains 0.3.0. Do not claim it supports this complete workflow.
+
+## Before paid work
+1. Read dreamlayer_capabilities and dreamlayer_balance (or CLI capabilities/balance --json). If either fails, resolve it before spending.
+2. Confirm the requested operation is advertised. Ordinary completed image operations currently cost one credit each; obtain the live cost from capabilities. A multi-step edit and cutout costs two operations.
+3. Use the user's already-authorized operation, number of outputs and credit budget. Ask only if missing. Never buy credits or expand the batch implicitly.
+4. For file operations, verify input readability and a new output path before submitting. Upload once using dreamlayer_upload_image; preserve its input_asset_id.
+5. Save a unique idempotency_key and the complete request locally before each generation. Do not store credentials in the job record.
+
+## Execute and recover
+Select operation explicitly: text_to_image; image_to_image; background_remove; upscale; sprite_sheet.
+All except text_to_image require input_asset_id. Generate returns execution_id, status, asset, question and last_event_id.
+Preserve IDs/cursor after each response. For running work, call dreamlayer_execution or dreamlayer_events; do not generate a replacement after a timeout.
+A needs_input result is a question, not a failed generation. Continue using conversation_id and respond.
+For completed work, dreamlayer_download writes an absolute new local path. Never treat a returned asset URL as proof that the local file exists.
+On a local download error, repair the destination and download the same execution.
+File-based CLI retries upload a new asset; use status and download after uncertainty rather than rerunning the paid command.
+Stop at insufficient credits, invalid request, unsupported operation, or an unresolved contract error. Keep recoverable state; do not loop paid retries.
+
+## Output acceptance
+Inspect the saved image or preview. Check dimensions, transparency if requested, text spelling, reference fidelity and the intended use. Label uninspected outputs.
+PNG output does not establish SVG/vector structure, pixel-perfect art, trademark clearance, print compliance, exact product fidelity or consistent animation.
+API documentation: https://docs.dreamlayer.io/agent-api
+Recovery: https://docs.dreamlayer.io/agent-api/jobs-and-events
+
+## CLI examples
+```sh
+dreamlayer capabilities --json
+dreamlayer balance --json
+dreamlayer generate "Original image brief" --aspect 1:1 --idempotency-key UNIQUE_SAVED_KEY --out concept.png --json
+dreamlayer edit reference.png "Change the background to a simple studio scene" --idempotency-key UNIQUE_SAVED_KEY --out edited.png --json
+dreamlayer cutout reference.png --idempotency-key UNIQUE_SAVED_KEY --out cutout.png --json
+dreamlayer status EXECUTION_ID --json
+dreamlayer download EXECUTION_ID --out recovered.png --json
+```
+Placeholder IDs and keys must be replaced. Check CLI --help for supported flags; --max-credits is for sprite jobs.
+