rsn-perceiving-information
npx skills add https://github.com/bellabe/lean-os --skill rsn-perceiving-information
Agent 安装分布
Skill 文档
Perceiving
Direct attention. Filter information. Feed thinking.
Core Principle
Perception precedes thought. What you attend to determines what you can reason about.
perceiving â thinking â output
â â
ââââââââââââââ
(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 â Scanning
NO â
Do you need deep understanding of something specific?
YES â Focusing
NO â
Are you overwhelmed with information?
YES â Filtering
NO â
Do you need to verify something from multiple sources?
YES â Triangulating
NO â
Do you need continuous awareness of a condition?
YES â Monitoring
NO â
Do you have multiple signals that need integration?
YES â Synthesizing
NO â No perception mode needed
Mode Summaries
Scanning
Purpose: Broad, shallow attention across a space.
Mental model: Radar sweepâcover 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
Focusing
Purpose: Deep, narrow attention on specific signal.
Mental model: Microscopeâsacrifice 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
Filtering
Purpose: Separate signal from noise.
Mental model: Coffee filterâlet 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
Triangulating
Purpose: Cross-reference multiple sources for verification.
Mental model: Navigationâfix position from multiple bearings.
Patterns: 3-Independent (strongest), Primary-Secondary, Multi-Method, Temporal
Key rules:
- Verify source independence (not echoes)
- Require â¥3 sources for high-stakes claims
- Investigate conflicts, don’t dismiss
- Adjust confidence based on agreement
Output: Verified claim with confidence and caveats
â references/triangulating.md
Monitoring
Purpose: Continuous watch for threshold or condition.
Mental model: Smoke detectorâconstant 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
Synthesizing
Purpose: Integrate multiple signals into coherent picture.
Mental model: Jigsaw puzzleâassemble 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
â references/synthesizing.md
Output Format
Prose, not YAML. Every perceiving output includes:
## [Mode]: [Topic]
**Summary:** [Key finding in 1-2 sentences]
**Confidence:** [X%] â [Why this confidence level]
**Sources:** [What was examined]
**Signals:**
- [Signal 1]: [Relevance] â [Action: focus/monitor/ignore]
- [Signal 2]: [Relevance] â [Action]
**Blind spots:** [What might be missed]
**Next:**
- Focus on: [Signals warranting deep dive]
- Monitor: [Signals warranting watch]
- Ready for thinking: [Yes/No â 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 â 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 |