Externalize AI system task prompts into versioned artifacts

[chubes4]

Problem

AI-backed system tasks may embed prompt behavior in PHP classes or hardcoded defaults. That makes task quality hard to update safely across installs.

Desired shape

System tasks that use AI prompts should load prompts from a versioned prompt artifact registry.

Needed pieces:

• system task prompt registry
• default prompt artifact IDs
• versioned prompt loading
• site/local override support
• installed/current hash tracking
• upgrade planner integration

Acceptance criteria

• AI system tasks can declare a default prompt artifact.
• The default prompt can be updated by platform/plugin releases.
• Untouched installed prompts auto-update.
• Locally modified prompts stage a PendingAction instead of being overwritten.
• Runtime execution can resolve the effective prompt deterministically.

Context

Examples include daily memory generation, wiki maintenance, meta description generation, alt text generation, summarization, and future graph extraction tasks.

AI assistance

• AI assistance: Yes
• Tool(s): OpenCode (GPT-5.5)
• Used for: Drafting the issue from product/design discussion; Chris remains responsible for prioritization and implementation.

[chubes4]

Partially shipped in Data Machine v0.97.1. PR #1549 added SystemTaskPromptRegistry and SystemTask::resolvePrompt(), so AI-backed SystemTasks can expose getPromptDefinitions(), derive stable system-task:: PromptArtifact IDs, carry version/source_path/changelog metadata, and resolve through the datamachine_system_task_effective_prompt filter. Daily memory, meta description, internal linking, alt text, and similar tasks now have the seam. What remains: prompts still live as defaults in task definitions, not external bundle files; installed/current hashes, local override storage, and upgrade-planner application for system task prompts are not fully wired. Leaving open for the externalized artifact-store/upgrade slice.