This skill is responsible for creating compositions, setting canvas frame rate duration, and workspace. It hands off to skills in the same package by name, can be used when installed alone, and does not read sibling directories. It invokes the fixed official effectcraft-cli and preserves the native editing project.
Input and delivery
The input is the user-confirmed task, materials, project or object, output directory, and modification scope; when an existing project is needed, verify the summary first. Return the actual CLI results, the saved native project, the required derived outputs, and the verification records. Report installation success, command directory existence, and creation task completion separately.
Default managed entry point
Execution implementation converges in the built-in effectcraft-harness skill (installed bundled with the plugin; for standalone installation use npx skills add full-aigc-skills/effectcraft-skills --skill effectcraft-harness). Set HARNESS_DIR to the directory where that skill is actually loaded (absolute path); this skill's SKILL_DIR is only used to read its own examples/ and references/。
Set from the skill directory actually loaded by the host SKILL_DIR。Creation writes use by default launch.sh / launch.ps1, automatically preparing the fixed Python and EffectCraft inside the user directory, without changing the global PATH. The raw Python/CLI examples below are retained for compatibility and diagnostic purposes and must not bypass the task recovery ledger. For the complete operation contract see Task control, evaluation, and revision。
/bin/sh "$HARNESS_DIR/scripts/launch.sh" plan --mode workflow --plan PLAN.json --output /absolute/new-result
/bin/sh "$HARNESS_DIR/scripts/launch.sh" run --mode workflow --plan PLAN.json --output /absolute/new-result
Windows PowerShell：& "$HARNESS_DIR/scripts/launch.ps1" run --mode workflow --plan PLAN.json --output "C:\Users\me\result".
Native command plan appends --mode commands, owned desktop sessions use --mode desktop. After delivery you must review and actually view the media; report as acceptable only when both technical verification and creative evaluation pass, and the user acceptance status is kept independently. When automatic revision is needed, declare it in the first plan revisionScope; by default at most 2 rounds, parent and child tasks share 30 minutes. Unknown tasks are only inspect/reconcile, do not redo with a different task ID.
First use and public entry point
Locate the real directory of the current SKILL.md. Native artifacts cover macOS arm64/x86_64, Windows x86/x64/arm64, Linux x86_64/aarch64; no preinstalled Python is required; the fixed CLI installs into the user data directory. When existing task authorization covers the necessary dependencies, run this skill's installer directly, and do not create a separate approval process.
Set SKILL_DIR to the directory containing this SKILL.md actually loaded by the host (absolute path). User-level installation may be located at ~/.agents/skills/effectcraft-cli-composition, project-level may be located at .agents/skills/effectcraft-cli-composition, the plugin may be located in its skills/effectcraft-cli-composition or host cache directory; use the actual load path as the standard, do not guess based on the current working directory, and do not search and arbitrarily select a duplicate version. The skill directory and the CLI's user data installation directory are two independent locations.
: "${SKILL_DIR:?请先设置为本 SKILL.md 的实际所在目录}"
python3 -I -B "$HARNESS_DIR/scripts/bootstrap.py"
python3 -I -B "$HARNESS_DIR/scripts/cli.py" -- --version
python3 -I -B "$HARNESS_DIR/scripts/cli.py" -- commands --json
CLI argv comes after -- , and the native subcommand must be first. Installation parameters go before the separator;--runtime-home can isolate the cache. Stop when locking the artifact digest fails, the installation is corrupted, or the platform is unsupported; do not change PATH and do not perform floating upgrades. The native help entry point is --help; the launcher's own --help only explains startup parameters.
Scene operations
Verify this scene's input, native project, target object, version, and output boundaries. According to the scene guide , select the current command, save an independent checkpoint, then execute; after completion, reopen the native project and inspect the actual output and non-target content.
commands --filter <关键词> --json verify parameters; for property read/write use props/get/set, for commands use exec <id> --params <JSON> or paired run <id> <JSON> ...; use --save-as to keep a new project.
The composite frame rate and duration come from the requirements; verify nested compositions, and ensure preview and final output use the same time range.
Native compositions and source project revisions use the effectcraft-harness skill's scripts/workflow.py(reused internally by the managed launcher); read the workflow contract, with templates in this skill's examples. Modify only authorized objects, preserve native projects and dependent assets, and read the derived format loss exchange report。
Verification and Recovery
A non-zero command exit does not count as completion; you need to inspect the actually saved project, export, dimensions, and media decoding. Mark timeout results as unknown, check the original task/project first, and do not automatically replay edits. The existing native workflow supports saving revisions; when the corresponding project summary does not match, refuse to overwrite.
On-demand Reference and Handoff
- Actual command evidence: fixed-version observations, used only as routing and parameter reference; live results take precedence, and disabled items are not executed.
- Workflow contract: composition operations, native saving, dependency collection, and revisions.
- When installation/diagnostics are needed, hand off to effectcraft-cli-setup, for full task routing hand off to effectcraft-use; when a skill is missing, use
npx skills add full-aigc-skills/effectcraft-skills --skill <skill-name>. Do not load other skills via adjacent file paths.
This skill does not provide a fictitious login interface; local headless does not require a cloud account. Full GUI, cross-editor fidelity, and creative quality are stated according to actual evidence.
Full native commands use
The query example below requires an existing Python; if no interpreter is prepared, first use the managed launcher doctor for diagnostics. The write example automatically prepares isolated dependencies through the launcher; on Windows use & "$HARNESS_DIR/scripts/launch.ps1" instead of /bin/sh "$HARNESS_DIR/scripts/launch.sh", keeping the same mode and parameters. The current skill includes a complete directory of parameter descriptions and same-session entry points, and is not restricted by the creative template allowlist. Read the complete usage guide, and query the command referenceas needed; each instruction has a skill route, prerequisite observations, and acceptance status.
python3 -I -B "$HARNESS_DIR/scripts/commands.py" list --filter QUERY
python3 -I -B "$HARNESS_DIR/scripts/commands.py" describe COMMAND_ID
/bin/sh "$HARNESS_DIR/scripts/launch.sh" run --mode commands --plan "$SKILL_DIR/examples/commands-advanced.json" --output /absolute/new-command-result
/bin/sh "$HARNESS_DIR/scripts/launch.sh" run --mode desktop --plan "$SKILL_DIR/examples/desktop-first-use.json" --output /absolute/new-desktop-result
Before executing a new entry point, check the real registry and current executable status, keep return value references and step-by-step receipts; semantic errors or timeouts must not be passed off as success. Directory coverage and direct native use do not equal acceptance of all instructions, GUI, delivery, or Art orchestration.
The complete workflow command gateway is in usage instructions. For fixed CLI versions and artifact digests, refer to the effectcraft-harness skill's scripts/runtime.lock.json ; skill package versions are based on the corresponding release tags. Independent installation re-verification and full command-by-command acceptance are recorded separately, and release is not a substitute for acceptance.
Parent-child layers, property expressions, dynamic preview, and source project rework: use the current skill's expression parenting guide。
For GUI tasks, you may first use this skill's built-in fixed desktop installation; installation, launch, and actual GUI editing are verified separately.
Complete attribution commands are listed by command family in the categorized usage list, and each item can be viewed via this skill's commands.py describe parameters; the general CLI list also retains all commands not assigned to specialized skills.
Business tasks start from the scenario operation manual , following input checks, template adaptation, native delivery, localized rework, and result verification.
When installation fails, read first-use diagnostics, locate the current skill's own setup entry point based on the receipt; handle installation failures and native invocation failures separately.
Read the relevant creative task shot tracking scenarioto verify native parameters, object context, local rework, and delivery boundaries.
Same-target execution protection
Before invoking the public workflow, read this skill's execution registration and interruption handling. In the event of contention or an unknown state, do not delete the registration, automatically replay, or bypass verification by switching targets.