SuperAPI AI Directory

insight

Writing & research

View repository

Updated 10-20 16:16

About this skill

Insight

Read between the lines to surface what someone means but is not saying.

What This Skill Does

In D&D, Insight detects lies and reads true intentions. The real-world version is subtext analysis: figuring out what an email actually means, what a stakeholder's real objection is, what a user's bug report is actually describing, or what the unsaid concern is behind a seemingly simple question. Insight does not read minds — it reads signals that are present but not foregrounded. In this grimoire, Insight is treated as a metaphorical skill with a shipping-now delivery profile. Canonical reference input: Insight (skill).

When To Use

  • A message, request, or conversation has an obvious surface meaning and a likely different real meaning.
  • You need to understand what someone actually wants rather than what they literally asked for.
  • A stakeholder's stated objection does not match their behavior, and you need to identify the real blocker.

Prerequisites

  • No extra runtime dependencies beyond Hermes Agent and the normal toolset for this session.

Procedure

  1. Restate the target, the success condition, and any no-touch boundaries before taking action.
  2. Read the surface communication: what is being explicitly said or asked.
  3. Identify the signals that do not fit the surface reading: tone shifts, conspicuous omissions, hedging language, over-specificity.
  4. Generate the most likely subtext: what is the real concern, objection, or request?
  5. Present both readings — surface and subtext — with confidence levels. Never claim the subtext is certain.
  6. Suggest a response that addresses the subtext without calling it out aggressively.
  7. Package the result as the deliverables below, with confidence, assumptions, and unresolved risk called out explicitly.

Deliverables

  • The surface reading: what was literally said.
  • The subtext reading: what is likely meant, with evidence from the signals.
  • A suggested response that addresses the real concern diplomatically.

Pitfalls / Guardrails

  • Keep the metaphor anchored to a real mechanism instead of drifting into lore.
  • Subtext analysis is inference, not mind-reading. Always present it as a hypothesis with supporting evidence, never as certainty.
  • Do not project malicious intent without strong signal. Most communication gaps are confusion, not manipulation.

Verification

  • Check that the result includes every deliverable promised above.
  • Check that confirmed facts, assumptions, and inferences are visibly separated.
  • Check that the metaphor still maps cleanly to a real operational mechanism.

Example Invocation

/insight read between the lines of this [message/conversation/request]. What is the surface meaning, what is the likely subtext, and how should I respond to the real concern?