rsn-perceiving-information

Direct attention. Filter information. Feed thinking.

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "rsn-perceiving-information" with this command: npx skills add bellabe/lean-os/bellabe-lean-os-rsn-perceiving-information

Perceiving

Direct attention. Filter information. Feed thinking.

Core Principle

Perception precedes thought. What you attend to determines what you can reason about.

perceiving → thinking → output ↑ │ └────────────┘ (need more data)

Bad perception = reasoning on wrong inputs = wrong conclusions.

Mode Selection

Mode Question Output Trigger

Scanning What's out there? Landscape summary + signals Broad awareness needed

Focusing What's the detail here? Deep understanding Specific signal important

Filtering What's noise vs signal? Prioritized list Information overload

Triangulating Do sources agree? Verified claim Verification needed

Monitoring Has threshold crossed? Status + alerts Continuous watch required

Synthesizing What's the pattern across inputs? Integrated picture Multiple signals need combining

Decision Tree

Do you need broad awareness of a space? YES → Scanning NO ↓ Do you need deep understanding of something specific? YES → Focusing NO ↓ Are you overwhelmed with information? YES → Filtering NO ↓ Do you need to verify something from multiple sources? YES → Triangulating NO ↓ Do you need continuous awareness of a condition? YES → Monitoring NO ↓ Do you have multiple signals that need integration? YES → Synthesizing NO → No perception mode needed

Mode Summaries

Scanning

Purpose: Broad, shallow attention across a space.

Mental model: Radar sweep—cover maximum area, detect anything notable.

Patterns: Sweep (everything once), Sample (representative subset), Edge (periphery/outliers), Competitive (key players only)

Key rules:

  • Bound scope and time (not infinite)

  • Note signals, don't investigate yet

  • Acknowledge blind spots

  • Triage: focus / monitor / ignore

Output: Landscape summary + prioritized signals

→ references/scanning.md

Focusing

Purpose: Deep, narrow attention on specific signal.

Mental model: Microscope—sacrifice breadth for depth.

Approaches: Exhaustive (all sources), Targeted (answer specific questions), Iterative (deepen until diminishing returns)

Key rules:

  • Define questions before diving in

  • State confidence per finding

  • Document unknowns

  • Extract implications for thinking/action

Output: Detailed understanding with implications

→ references/focusing.md

Filtering

Purpose: Separate signal from noise.

Mental model: Coffee filter—let through what matters, block the rest.

Filter types: Threshold, Categorical, Recency, Source, Relevance

Key rules:

  • Define criteria before filtering

  • Sample filtered-out items (check for false negatives)

  • Assess false negative risk

  • Adjust criteria if needed

Output: Prioritized list with filtering rationale

→ references/filtering.md

Triangulating

Purpose: Cross-reference multiple sources for verification.

Mental model: Navigation—fix position from multiple bearings.

Patterns: 3-Independent (strongest), Primary-Secondary, Multi-Method, Temporal

Key rules:

  • Verify source independence (not echoes)

  • Require ≥3 sources for high-stakes claims

  • Investigate conflicts, don't dismiss

  • Adjust confidence based on agreement

Output: Verified claim with confidence and caveats

→ references/triangulating.md

Monitoring

Purpose: Continuous watch for threshold or condition.

Mental model: Smoke detector—constant vigilance, alert on trigger.

Types: Threshold (value crosses line), Trend (direction changes), Anomaly (deviation from normal), Absence (expected event missing), Combination (multiple conditions)

Key rules:

  • Quantify thresholds

  • Define severity levels

  • Specify response actions

  • Tune to prevent alert fatigue

Output: Status report with any alerts

→ references/monitoring.md

Synthesizing

Purpose: Integrate multiple signals into coherent picture.

Mental model: Jigsaw puzzle—assemble pieces into whole.

Methods: Weighted Average (quantitative), Bayesian (probabilistic), Narrative (qualitative story), Framework (structured template)

Key rules:

  • Weight inputs by quality/reliability

  • Resolve or flag conflicts

  • Verify coherence (does synthesis explain all signals?)

  • Make testable predictions

Output: Integrated picture ready for thinking

→ references/synthesizing.md

Output Format

Prose, not YAML. Every perceiving output includes:

[Mode]: [Topic]

Summary: [Key finding in 1-2 sentences]

Confidence: [X%] — [Why this confidence level]

Sources: [What was examined]

Signals:

  • [Signal 1]: [Relevance] — [Action: focus/monitor/ignore]
  • [Signal 2]: [Relevance] — [Action]

Blind spots: [What might be missed]

Next:

  • Focus on: [Signals warranting deep dive]
  • Monitor: [Signals warranting watch]
  • Ready for thinking: [Yes/No — which mode suggested]

Mode Transitions

From To Trigger

Scanning Focusing Signal detected worth deep dive

Scanning Monitoring Signal detected worth watching

Focusing Triangulating Finding needs verification

Monitoring Focusing Threshold crossed, need understanding

Any Filtering Information overload

Multiple Synthesizing Multiple signals need integration

Perceiving → Thinking Handoff

Perceiving Output Thinking Mode

Synthesized landscape Inductive (find patterns)

Verified anomaly Abductive (diagnose cause)

Conflicting signals Dialectical (resolve conflict)

Deep understanding of option Counterfactual (evaluate)

Filtered priorities Causal (plan actions)

Anti-Patterns

Avoid Do Instead

Boiling the ocean Bound scope, time-limit scanning

Rabbit holes during scanning Note and move on

Confirmation bias Include disconfirming sources

Single source trust Triangulate important claims

Alert fatigue Tune thresholds, prioritize severity

Analysis paralysis Set criteria, move forward

References

File Content

scanning.md Survey patterns and landscape mapping

focusing.md Deep dive methods

filtering.md Noise reduction techniques

triangulating.md Verification patterns

monitoring.md Alert configuration

synthesizing.md Integration methods

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

General

rsn-learning-outcomes

No summary provided by upstream source.

Repository SourceNeeds Review
General

rsn-creating-ideas

No summary provided by upstream source.

Repository SourceNeeds Review
General

sys-activating-goals

No summary provided by upstream source.

Repository SourceNeeds Review
General

sys-decomposing-goals

No summary provided by upstream source.

Repository SourceNeeds Review