AITP Runtime Mode Selector

Environment gate

• This skill runs inside an AITP session after research mode and load profile have been classified.
• Load this skill when aitp_run_topic_loop or aitp_resume_topic is about to select the runtime mode.

Classification task

Runtime mode

Choose between three runtime modes:

Mode	When to choose
explore	The topic is gathering information, surveying literature, or brainstorming. No bounded target yet; the goal is to map the landscape.
learn	The topic has a bounded question and is building understanding toward answering it. Reading papers, running initial calculations, constructing mental models.
implement	The topic has a clear plan and is executing: running calculations, writing derivations, producing artifacts, or conducting validation rounds.

Active submode

Choose between six submodes (only one active at a time):

Submode	When to choose
derivation	Active formal derivation work: proving theorems, expanding identities, closing proof obligations.
numerical	Active computation: running ab initio codes, solving lattice models, performing numerical analysis.
literature	Active literature intake: reading papers, extracting claims, building citation networks.
code	Active code development: writing scripts, fixing bugs, building tooling.
formal	Active formalization work: translating informal proofs to formal systems, checking type theory, Lean proofs.
experimental	Active experimental work: setting up and running physical or computational experiments.

Reasoning priority

1. Topic state: Check topic_state.json for the current phase. Early phases → explore. Middle → learn. Late with clear plan → implement.
2. Research mode signal: formal_derivation → likely implement with derivation or formal submode. first_principles → likely implement with numerical submode. exploratory_general → likely explore or learn.
3. Human request: What did the user just ask for? "Run the calculation" → implement + numerical. "Read this paper" → learn + literature. "I think the proof goes like this" → implement + derivation.
4. Transition posture: If the topic just entered a new phase (e.g., from intake to execution), upgrade the mode accordingly. The transition rules in mode_envelope_data.json provide deterministic guidance for allowed transitions.
5. Default: explore with no submode.

Recording the classification

Record both mode and submode in one call each:

aitp_record_classification(
    topic_slug=<current topic>,
    classification_type="runtime_mode",
    value=<"explore" or "learn" or "implement">,
    rationale=<1-2 sentence explanation>,
    signals_used=<list of signals>,
    source="ai_reasoning"
)

aitp_record_classification(
    topic_slug=<current topic>,
    classification_type="active_submode",
    value=<submode string or "none">,
    rationale=<1-2 sentence explanation>,
    signals_used=<list of signals>,
    source="ai_reasoning"
)

Hard rules

• Do not use keyword or substring matching on the human request.
• Mode transitions must respect the deterministic rules in mode_envelope_support.py (forbidden shortcuts, layer permissions). These are mechanical checks, not AI decisions.
• Record both mode and submode before the loop step executes.
• If the topic is still in an early phase with no clear direction, prefer explore with no submode.