Start here for every task.

Lists active task contexts (reusable task TYPES, not task instances).

Next steps:

• If a context matches: call get_artifacts_for_task_context(task_context_id)

• If no context matches: call create_task_context(summary, description)

Create a new task context (task type) when no match exists.

Use for categories (e.g., "CV analysis for Python dev"), not specific instances.

Constraints:

• English only

• summary <= 200 chars

• description <= 1000 chars

Next step: create initial guidance with create_artifact() before doing task work.

Load artifacts for a task context.

Call this after you select or create a task context and before doing work. Re-call when you start a new phase or need to confirm guidance.

Notes:

• Defaults to practice/rule/prompt (excludes result)

• Set include_archived=True only when you need historical context

Create a new artifact to capture reusable guidance.

Create immediately when you discover a pattern, constraint, mistake, or useful template. If similar guidance might already exist, call search_artifacts() first; prefer update_artifact() over near-duplicates.

Constraints:

• English only

• summary <= 200 chars

• content <= 4000 chars

• No PII, no task-instance specifics; focus on WHAT/WHY

Types: practice (guidelines), rule (constraints), prompt (templates), result (generalizable learnings).

Update an artifact when existing guidance is incomplete, wrong, or needs refinement.

Use immediately when you learn something better or user feedback indicates a correction. Prefer updating over creating duplicates.

Constraints:

• English only

• summary <= 200 chars

• content <= 4000 chars

• No PII, no task-instance specifics; focus on WHAT/WHY

Provide summary and/or content.

Archive an artifact that is incorrect, misleading, or outdated.

Prefer creating a replacement first, then archive the old artifact. Provide a reason when possible.

Full-text search across artifacts.

Use this before creating new artifacts to avoid duplicates. Returns results ranked by relevance.

Reflection checkpoint.

Call before declaring a task complete, and after corrections or user feedback. This returns the current artifacts and prompts you to create/update/archive as needed.