Workflow Orchestrator
Use this skill to route the active workflow, maintain visible state, assign bounded roles, enforce gates, and decide loopbacks. It owns workflow state, not specialist deliverables.
Native personal Codex Skills installed through $skill-installer do not require
project config, rendered agents or .framecore/manifest.json. Resolve the actual
installed Skill paths and keep the Workflow Profile and state visible; save
project files only with approval. Role IDs are task responsibilities unless the
host actually exposes the corresponding registered agents. Use the installed
pipeline-core role-to-skill map for supporting Skills, not invented role-named
Skill directories. Report unavailable dependencies; do not silently install them.
In a project-local Codex install, roles may resolve to rendered .codex/agents/*.toml. In ChatGPT, treat the same role IDs as temporary responsibilities inside the current task. Do not claim that ChatGPT created permanent agents or local project files.
Language Policy
Public instructions and installation guidance stay in English. After verified installation, resolve the working language from an explicit user preference, then the user's own conversation, then a reliably exposed host locale, otherwise English. Copied English setup prompts and source files are not language preferences; do not infer hidden account settings. Keep explicit deliverable-language requests and exact supplied artwork copy separate from conversation language. Do not translate or rewrite public source files to localize a session.
When To Use
Use this skill when:
- A request needs routing across multiple roles, skills, artifacts, or gates.
- The workflow has missing inputs, competing next steps, or possible loopbacks.
- The user asks what should happen next in the kit.
Do not use this skill to write specialist deliverables when a specialist role owns them.
Inputs
Required:
task_confirmation: confirmed goal, excluded scope, work mode, expected output, and immediate next step.
available_artifacts: brief, references, direction, prompts, QA, manifests, or docs already present.
user_request: current instruction and any new constraints.
Optional:
local_preferences: working language, tone, QA strictness, display names, output path.
repo_state: changed files, test status, issue list, or validation results when a repository is actually available.
delivery_request: only when explicit in the current user message.
Outputs
Produce Project State with:
- workflow blueprint
- active roles and responsibilities
- completed or missing artifacts
- last completed gate
- required handoffs
- review gates
- pending decisions and blocked items
- files touched and visible risks
- loopback decisions
- next role and next action
- recovery prompt for context loss
Process
- Start from
intent-confirmation.
- Inspect which artifacts already exist and which gates are blocked.
- Select the closest workflow blueprint when the task matches a common route.
- Shrink or expand the route based on available artifacts and explicit user scope.
- Assign specialist work without taking over specialist outputs. In ChatGPT, keep each assignment temporary and visible in the conversation.
- Decide loopback or delivery only after gates are satisfied.
Decision Rules
- Prefer the smallest sufficient route. If one specialist skill can produce the requested bounded artifact from available inputs, route directly to it after intent confirmation.
- Expand to a multi-stage pipeline only when the user explicitly requests end-to-end work or the task needs dependent artifacts, gates, handoffs, shared state, or QA.
- An explicit
$workflow-orchestrator invocation requests visible route selection and state tracking; it does not imply that every workflow stage must run.
- If the goal is unclear, route back to
intent-confirmation.
- If the brief is missing, route to
brief-architect.
- If references are required but unresolved, route to
reference-curator.
- If outputs exist, route to
qa-iteration before delivery.
- Use
instruction-packet-factory when delegation needs a bounded packet.
- Use
ecommerce-campaign-strategy-director when product, offer, channel, claim, asset-matrix, or test strategy is unresolved.
- Use
screenplay-story-architect before storyboard work when narrative structure, treatment, scenes, dialogue, pitch, or rewrite quality is the real blocker.
- Use
creative-video-producer when a video request spans several production layers instead of one specialist output.
- Use
caption-studio for detailed subtitle timing, styling, safe zones, or caption QA, and opencut-video-studio for footage-first timeline planning.
- Use
remotion-video-production when the deliverable is a deterministic React/TypeScript video composition, reusable template, or data-driven render plan.
- Use
producer-ai-task-builder for text-only music and music-video execution packets while keeping provider execution outside the route.
- Use workflow blueprints as defaults, not as rigid scripts.
Guardrails
- Do not skip review gates for speed.
- Do not route directly from prompts to delivery when generated assets exist.
- Do not enable external execution, upload, or publishing without explicit user instruction.
- Do not import private workspace context into public artifacts.
- Do not mutate workflow rules without explicit approval.
- Do not claim local files, shell checks, persistent agents, or repository state on a ChatGPT-only surface.
Handoff
Review gate: workflow_route.
Hand off with:
workflow_blueprint
active_roles
completed_or_existing_artifacts
last_completed_gate
required_handoffs
review_gates
pending_decisions
blocked_items
files_touched
risks
next_role
next_action
recovery_prompt
loopback_reason
QA Checklist
- Route follows the current user request.
- Route depth is justified, and a simple request does not expand into an unnecessary full pipeline.
- Required artifacts and missing blockers are visible.
- Each active role has a clear output.
- Gates and handoffs match the pipeline references.
- Next action is concrete and safe.
1---2name: workflow-orchestrator3description: Use this skill to route multi-stage work in Codex or ChatGPT, maintain visible workflow state, assign bounded roles, enforce gates, and decide loopbacks.4---56# Workflow Orchestrator78Use this skill to route the active workflow, maintain visible state, assign bounded roles, enforce gates, and decide loopbacks. It owns workflow state, not specialist deliverables.910Native personal Codex Skills installed through `$skill-installer` do not require11project config, rendered agents or `.framecore/manifest.json`. Resolve the actual12installed Skill paths and keep the Workflow Profile and state visible; save13project files only with approval. Role IDs are task responsibilities unless the14host actually exposes the corresponding registered agents. Use the installed15`pipeline-core` role-to-skill map for supporting Skills, not invented role-named16Skill directories. Report unavailable dependencies; do not silently install them.1718In a project-local Codex install, roles may resolve to rendered `.codex/agents/*.toml`. In ChatGPT, treat the same role IDs as temporary responsibilities inside the current task. Do not claim that ChatGPT created permanent agents or local project files.1920## Language Policy2122Public instructions and installation guidance stay in English. After verified installation, resolve the working language from an explicit user preference, then the user's own conversation, then a reliably exposed host locale, otherwise English. Copied English setup prompts and source files are not language preferences; do not infer hidden account settings. Keep explicit deliverable-language requests and exact supplied artwork copy separate from conversation language. Do not translate or rewrite public source files to localize a session.2324## When To Use2526Use this skill when:2728- A request needs routing across multiple roles, skills, artifacts, or gates.29- The workflow has missing inputs, competing next steps, or possible loopbacks.30- The user asks what should happen next in the kit.3132Do not use this skill to write specialist deliverables when a specialist role owns them.3334## Inputs3536Required:3738- `task_confirmation`: confirmed goal, excluded scope, work mode, expected output, and immediate next step.39- `available_artifacts`: brief, references, direction, prompts, QA, manifests, or docs already present.40- `user_request`: current instruction and any new constraints.4142Optional:4344- `local_preferences`: working language, tone, QA strictness, display names, output path.45- `repo_state`: changed files, test status, issue list, or validation results when a repository is actually available.46- `delivery_request`: only when explicit in the current user message.4748## Outputs4950Produce Project State with:5152- workflow blueprint53- active roles and responsibilities54- completed or missing artifacts55- last completed gate56- required handoffs57- review gates58- pending decisions and blocked items59- files touched and visible risks60- loopback decisions61- next role and next action62- recovery prompt for context loss6364## Process65661. Start from `intent-confirmation`.672. Inspect which artifacts already exist and which gates are blocked.683. Select the closest workflow blueprint when the task matches a common route.694. Shrink or expand the route based on available artifacts and explicit user scope.705. Assign specialist work without taking over specialist outputs. In ChatGPT, keep each assignment temporary and visible in the conversation.716. Decide loopback or delivery only after gates are satisfied.7273## Decision Rules7475- Prefer the smallest sufficient route. If one specialist skill can produce the requested bounded artifact from available inputs, route directly to it after intent confirmation.76- Expand to a multi-stage pipeline only when the user explicitly requests end-to-end work or the task needs dependent artifacts, gates, handoffs, shared state, or QA.77- An explicit `$workflow-orchestrator` invocation requests visible route selection and state tracking; it does not imply that every workflow stage must run.78- If the goal is unclear, route back to `intent-confirmation`.79- If the brief is missing, route to `brief-architect`.80- If references are required but unresolved, route to `reference-curator`.81- If outputs exist, route to `qa-iteration` before delivery.82- Use `instruction-packet-factory` when delegation needs a bounded packet.83- Use `ecommerce-campaign-strategy-director` when product, offer, channel, claim, asset-matrix, or test strategy is unresolved.84- Use `screenplay-story-architect` before storyboard work when narrative structure, treatment, scenes, dialogue, pitch, or rewrite quality is the real blocker.85- Use `creative-video-producer` when a video request spans several production layers instead of one specialist output.86- Use `caption-studio` for detailed subtitle timing, styling, safe zones, or caption QA, and `opencut-video-studio` for footage-first timeline planning.87- Use `remotion-video-production` when the deliverable is a deterministic React/TypeScript video composition, reusable template, or data-driven render plan.88- Use `producer-ai-task-builder` for text-only music and music-video execution packets while keeping provider execution outside the route.89- Use workflow blueprints as defaults, not as rigid scripts.9091## Guardrails9293- Do not skip review gates for speed.94- Do not route directly from prompts to delivery when generated assets exist.95- Do not enable external execution, upload, or publishing without explicit user instruction.96- Do not import private workspace context into public artifacts.97- Do not mutate workflow rules without explicit approval.98- Do not claim local files, shell checks, persistent agents, or repository state on a ChatGPT-only surface.99100## Handoff101102Review gate: `workflow_route`.103104Hand off with:105106- `workflow_blueprint`107- `active_roles`108- `completed_or_existing_artifacts`109- `last_completed_gate`110- `required_handoffs`111- `review_gates`112- `pending_decisions`113- `blocked_items`114- `files_touched`115- `risks`116- `next_role`117- `next_action`118- `recovery_prompt`119- `loopback_reason`120121## QA Checklist122123- Route follows the current user request.124- Route depth is justified, and a simple request does not expand into an unnecessary full pipeline.125- Required artifacts and missing blockers are visible.126- Each active role has a clear output.127- Gates and handoffs match the pipeline references.128- Next action is concrete and safe.