Socratic Grounding

Scope

• In-scope: drift checks, calibration dialogue, grounding against the local operating context, and closing a discussion into a short canonical rule.
• Out-of-scope: replacing execution discipline, doing broad planning when the task is still concrete, or turning a simple operational rule into a large taxonomy/framework exercise.

Indexed Reference Map

• @[SKILL.md#L1-L55] Frontmatter + complete local-ready skill definition.
• @[SKILL.md#L12-L18] Scope and intended use boundaries.
• @[SKILL.md#L20-L25] Trigger conditions and positive indicators.
• @[SKILL.md#L27-L41] SOP for drift check, grounding dialogue, and canon commit.
• @[SKILL.md#L43-L55] Error handling and output contract.

When to use

• When Boss asks direct calibration questions such as: “am I over-complicating this?”, “do you understand the environment yet?”, “what is the real point here?”, or “close this topic with a canon rule”.
• When the agent is drifting into taxonomy, wording refinement, abstraction, or generic assumptions instead of the current operational problem.
• When a decision has become clear enough that it should be compressed into a reusable canonical rule.

SOP

1. Drift Check
   • Answer the core question directly: Yes or No when asked whether the current reasoning is over-engineered, mis-grounded, or off-target.
   • If Yes, acknowledge it immediately and identify the excess layer.
   • If No, justify briefly based on operational value rather than rhetorical elegance.
2. Grounding Dialogue
   • Re-anchor on 3 axes:
     • Nature: what is actually being built, decided, or optimized?
     • Environment: what local contract or runtime facts control this case?
     • Logic: what must be kept, what is excess, and where is the stopping point?
   • Prefer local contract over generic assumptions.
   • Keep the answer compact and non-defensive.
3. Canon Commit
   • If Boss confirms the conclusion should be locked in, write a short canonical rule.
   • The rule should state: the base concept, the role boundary, any minimal classifier set that is truly needed, and the condition under which the rule should be revised.
   • Do not produce a canon rule longer or more complex than the actual decision requires.

Trigger Rules

• Positive: Trigger when the request matches socratic-grounding per the description: detect reasoning drift, calibrate against operational reality, answer direct meta-reasoning questions, or close a topic into a concise canon rule.
• Negative: Do not trigger when the task simply needs normal execution, routine debugging, or a concrete implementation workflow without any sign of reasoning drift.
• Escalation: If the discussion reveals a real policy/safety/runtime issue, hand back to the domain skill that owns the actual fix after the grounding step.

Error Handling

• If the calibration question is too vague: ask for the exact decision or reasoning branch to ground.
• If the answer depends on local runtime facts that are not yet verified: say so clearly and switch to evidence gathering instead of improvising.
• If the conversation is spiraling into wording/taxonomy refinement without operational value: stop and compress to the smallest useful canon rule or explicit stopping point.

Output Contract

• Deliverable: a concise calibration answer and, when requested, a short Canonical rule: block.
• Validation: the response must clearly separate what is essential, what is excess, and which local facts/control boundaries matter.
• Failure mode: if grounding cannot be completed safely, say what is missing and what evidence is required next.