From agent-almanac
Conducts structured deep listening to extract user intent from ambiguous requests, parsing literal, emotional, contextual layers, reflecting understanding, and noting unsaid elements before acting.
npx claudepluginhub pjt222/agent-almanacThis skill uses the workspace's default tool permissions.
---
Clarifies intent-expression gaps in user prompts via Hermeneia protocol dialogue, detecting ambiguities and extracting precise ClarifiedIntent through structured phases.
Guides developing active listening skills: mindset preparation, full-presence attending, paraphrasing, clarifying questions, synthesis for better communication and difficult conversations.
Adapts response style to user signals: relational clarification for hedging, open-ended, personal context; direct for transactional imperatives. Triggers on ambiguity in any conversation.
Share bugs, ideas, or general feedback.
Conduct a structured deep listening session — clearing assumptions, attending with full reception, parsing multiple signal layers, reflecting understanding back, noticing what is unsaid, and integrating the complete picture of the user's intent.
meditate clears internal noise, listen directs cleared attention outward toward the userBefore receiving the user's signal, release preconceptions about what they want.
Expected: A receptive state where attention is open rather than already narrowing toward a solution. The impulse to immediately respond is paused in favor of fully receiving.
On failure: If assumptions cannot be released (a strong pattern match persists), acknowledge the match explicitly: "This looks like X — but let me check if that is actually what is being asked." Naming the assumption weakens its grip.
Read the user's message with complete attention, holding all parts in awareness simultaneously.
Expected: A complete reception of the message — no words skipped, no sentences glossed over. The message is held as a whole rather than immediately decomposed into actionable parts.
On failure: If the message is very long, break it into sections but still read each section completely. If attention is pulled toward one part (usually the most technical), deliberately attend to the parts that are not technical — they often contain the intent.
The user's message contains multiple simultaneous signals. Parse each layer separately.
Signal Layer Taxonomy:
┌──────────────┬──────────────────────────────┬──────────────────────────┐
│ Layer │ What to Extract │ Evidence │
├──────────────┼──────────────────────────────┼──────────────────────────┤
│ Literal │ What the words explicitly │ Direct statements, │
│ │ say — the surface request │ specific instructions │
├──────────────┼──────────────────────────────┼──────────────────────────┤
│ Procedural │ What they want done — the │ Verbs, action words, │
│ │ desired action or output │ "I want," "please," │
│ │ │ "can you" │
├──────────────┼──────────────────────────────┼──────────────────────────┤
│ Emotional │ How they feel about the │ Frustration ("I keep │
│ │ situation — frustration, │ trying"), urgency ("I │
│ │ curiosity, urgency, delight │ need this now"), delight │
│ │ │ ("this is cool") │
├──────────────┼──────────────────────────────┼──────────────────────────┤
│ Contextual │ The situation surrounding │ Mentions of deadlines, │
│ │ the request — why now, │ other people, projects, │
│ │ what prompted it │ prior attempts │
├──────────────┼──────────────────────────────┼──────────────────────────┤
│ Constraint │ Boundaries on the solution │ "Without changing X," │
│ │ — what must be preserved, │ "keep it simple," │
│ │ what cannot change │ "compatible with Y" │
├──────────────┼──────────────────────────────┼──────────────────────────┤
│ Meta │ The request about the │ "Am I asking the right │
│ │ request — are they asking │ question?", "Is this │
│ │ whether they are asking │ even possible?", │
│ │ the right thing? │ "Should I be doing X?" │
└──────────────┴──────────────────────────────┴──────────────────────────┘
For each layer, note what is present and what is absent. The absent layers are as informative as the present ones.
Expected: A multi-layered reading of the message. The literal and procedural layers are usually clear. The emotional, contextual, constraint, and meta layers require more careful attention. At least one non-literal layer should be identified.
On failure: If only the literal layer is visible, the message may genuinely be straightforward — not all communication is layered. But check: is the message unusually short for its complexity? Are there hedging words ("maybe," "I think," "if possible")? These often indicate an unstated layer.
Before acting, reflect back what was heard to verify alignment.
Expected: The user confirms the reflection or corrects it. Either outcome is valuable — confirmation means the intent is aligned; correction means the intent is now clearer. The reflection should feel like a mirror, not a judgment.
On failure: If the user seems impatient with the reflection ("just do it"), they may value speed over alignment — honor that preference but note the risk of misalignment. If the reflection was wrong, do not defend it — accept the correction and update understanding immediately.
Attend to what the user did not say, which can be as important as what they did say.
Expected: At least one significant gap identified. This gap may not need to be addressed — but awareness of it prevents blind spots. The most useful gaps are missing constraints (the user assumed something they did not state) and missing context (why they need this now).
On failure: If no gaps are apparent, the user may have been thorough — but more likely, the gaps are in areas the AI is also blind to. Consider: what would a different person working on this project want to know that the user has not stated? This lateral perspective often surfaces hidden gaps.
Combine all layers and gaps into a unified picture of the user's actual need.
Expected: A complete, nuanced understanding of the user's need that goes beyond the surface request. The understanding is specific enough to guide action and honest enough to acknowledge uncertainty.
On failure: If integration produces a confused picture, the signals may genuinely conflict. In that case, ask one focused question that would resolve the ambiguity: "The most important thing for me to understand is..." Do not ask multiple questions — a single well-chosen question reveals more than a list of clarifications.
listen-guidance — the human-guidance variant for coaching a person in developing active listening skillsobserve — sustained neutral pattern recognition that feeds listening with broader contextteach — effective teaching requires listening first to understand the learner's needsmeditate — inward attention that clears the space for outward listeningheal — self-assessment that reveals whether the AI's listening capacity is impaired by drift