# Deep Creation Automation Boundary

```yaml
status: analysis-draft
canonical: false
created: 2026-06-15
scope: ccpe-system / writing-workbench / knowledge-vault
classification: boundary-analysis
```

## 1. Purpose

This draft records the current boundary for deep creation workflows.

It is not a canonical Runtime. It is an analysis document for later protocol updates.

## 2. Core Judgment

Deep creation can be supported by CCPE, but it should not be prematurely automated.

The user's earlier Web-based workflow worked because the user acted as the implicit runtime:

```text
choosing which expert to invoke
judging when context was polluted
deciding which critique mattered
deciding when to revise, stop, restart, or write
holding the final responsibility for taste, stance, and publication
```

When moved into Codex or another agentic environment, those hidden decisions must become explicit gates, records, and invocation boundaries. That is a governance problem, not a simple automation problem.

## 3. Not Automatically Automatable

Do not default to automating:

```text
article premise selection
whether an idea is worth writing
final thesis judgment
which review comments to accept
whether a structure carries the intended argument
whether a reviewer is producing useful tension or style noise
when to restart context
when a model should be promoted
when to enter full drafting
whether to publish
```

These are high-uncertainty judgment tasks. Agents may advise; the user directs.

## 4. Suitable for Context Assistance

CCPE may support:

```text
source digest drafting
source map preparation
premise context pack preparation
outline-aligned context pack preparation
role-specific dispatch packet drafting
material gap identification
conflict summary for user decision
repair instruction drafting after user decision
outline versioning rules
returned-output record templates
```

These tasks prepare context and preserve traceability. They do not replace human judgment.

## 5. Suitable for Automation

Automation is safer when the task is stable, low-risk, and verifiable:

```text
file naming
markdown heading repair
format conversion
template expansion
batch export
asset manifest generation
TTS batch preparation
image prompt list export
metadata formatting
archive packaging
```

These should usually belong in `skills-vault` as automation Skill source, with CCPE registration only when a CCPE Agent or Runtime depends on them.

## 6. Formal Participant Boundary

Canonical writer and reviewer output must come from real invocation.

Examples:

```text
Han Yu lead writer output
Zhang Liao red-team review
Giant Cognition macro-model review
Cognitive Imaging inspection
```

The controller session may prepare context and synthesize returned outputs. It must not impersonate the formal participant.

If invocation is not available, the workflow should produce:

```text
agent-invocation-packet.md
prompt-to-send.md
blocked_waiting_for_participant_output
```

## 7. Current Writing Workflow Boundary

For `writing-workbench`, the current expected flow is narrower:

```text
Web discussion and source digestion
-> Web initial outline v1.0
-> Writing Workbench project
-> outline review preparation
-> user-named reviewer invocation
-> user reflection and decision
-> user-directed lead writer repair
-> revised outline v2.0+
-> user-directed section / chapter drafting
```

The user decides every stage transition.

## 8. CCPE Supplier Role

CCPE should supply:

```text
lead writer Lite prompts or Agent specs
reviewer Lite prompts or Agent specs
invocation packet protocols
returned-output record templates
review report contracts
context preparation rules
source fidelity rules
evaluation rubrics
repair instruction builder rules
```

CCPE should not store one article's project records, reviewer sessions, draft chapters, or decision logs.

## 9. Stop Rule

Stop once the smallest useful artifact has been produced.

For deep creation, useful stopping points include:

```text
context pack ready for user review
prompt-to-send ready for real participant invocation
real participant output returned and recorded
conflict summary ready for user decision
repair instruction ready after user decision
revised outline returned and versioned
```

Do not continue into the next creative stage merely because another artifact can be generated.

## 10. Open Follow-Up

Future CCPE work may turn this draft into:

```text
ccpe-protocol deep creation boundary rule
writing-workbench supplier intake template
outline-review-and-repair runtime spec
reviewer invocation packet template
lead writer repair packet template
```

Promotion should happen only after concrete project requirements exist.