Upload plugins/ml-intern/skills/ml-intern/SKILL.md

plugins/ml-intern/skills/ml-intern/SKILL.md

@@ -39,3 +39,19 @@ For focused tasks, use the specialized skills directly:
 - `hf-paper-search`
 - `hf-docs`
 - `hf-jobs`
+
+## Re-engagement Rule
+
+Once `ml-intern-harness` (or this `ml-intern` alias) has been triggered in a conversation, the harness instructions must remain the active system instructions for the remainder of the session. Do not revert to generic assistant behavior unless the user explicitly says "stop using ml-intern" or the task is clearly non-ML.
+
+Before every response in a conversation where the harness has been active, check the user's last message for ML-related intent. If ANY of the following are true, re-invoke the harness workflow automatically (do not wait for the user to say "ml-intern" again):
+
+- The user requests training, fine-tuning, evaluation, dataset preparation, model cards, or benchmarking.
+- The user asks for a "plan", "architecture", "design", or "research" for an AI/ML system.
+- The user references a Hugging Face model, dataset, paper, or Space.
+- The user mentions: Hugging Face, HF, training, fine-tune, benchmark, RAG, embedding, diffusion, LoRA, DPO, GRPO, SFT, TRL, transformers, trackio, dataset, model, evaluate, inspect.
+- The prior turn in this conversation involved `ml-intern-harness`, `ml-intern`, or any sub-skill (`hf-paper-search`, `hf-dataset-search`, `hf-model-search`, `hf-docs`, `hf-jobs`, `github-example-search`).
+
+When re-engaging mid-conversation, do NOT restart from scratch. Resume from the most recently completed harness phase, read the conversation history for prior plan state and evidence, and proceed to the next logical step.
+
+If re-engagement is triggered, begin the response by stating the current harness phase (e.g., "Continuing from Phase 3: Implementation").