MCP HubMCP Hub
Volver a habilidades

observe

pjt222
Actualizado 2 days ago
5 vistas
17
2
17
Ver en GitHub
Metaai

Acerca de

La habilidad `observe` permite el monitoreo sistemático y pasivo de sistemas para identificar patrones sin intervención inmediata. Aplica una metodología de estudio naturalista—observar, registrar e hipotetizar—para comprender comportamientos poco claros o causas raíz. Úsala para depurar problemas desconocidos, evaluar cambios en el código o auditar tu propio razonamiento en busca de sesgos antes de tomar acción.

Instalación rápida

Claude Code

Recomendado
Principal
npx skills add pjt222/agent-almanac -a claude-code
Comando PluginAlternativo
/plugin add https://github.com/pjt222/agent-almanac
Git CloneAlternativo
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/observe

Copia y pega este comando en Claude Code para instalar esta habilidad

Documentación

Observe

Frame → witness → record → categorize → theorize → archive.

Use When

  • Behavior unclear → action premature
  • Debug unknown cause → observe before intervene → no symptom mask
  • Post-change → witness effects before more changes
  • User patterns over conv → improve future
  • Audit own reasoning → biases, habits, errors
  • After learn → validate model

In

  • Required: Target — system, codebase, behavior, user, reasoning
  • Optional: Duration/scope
  • Optional: Guiding question/hypothesis
  • Optional: Prior obs to compare (delta)

Do

Step 1: Frame

Define what + why + perspective.

Observation Protocol by System Type:
┌──────────────────┬──────────────────────────┬──────────────────────────┐
│ System Type      │ What to Observe          │ Categories to Watch      │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ Codebase         │ File structure, naming   │ Patterns, anti-patterns, │
│                  │ conventions, dependency  │ consistency, dead code,  │
│                  │ flow, test coverage,     │ documentation quality,   │
│                  │ error handling patterns  │ coupling between modules │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ User behavior    │ Question patterns,       │ Expertise signals, pain  │
│                  │ vocabulary evolution,    │ points, unstated needs,  │
│                  │ repeated requests,       │ learning trajectory,     │
│                  │ emotional signals        │ communication style      │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ Tool / API       │ Response patterns, error │ Rate limits, edge cases, │
│                  │ conditions, latency,     │ undocumented behavior,   │
│                  │ output format variations │ state dependencies       │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ Own reasoning    │ Decision patterns, tool  │ Biases, habits, blind    │
│                  │ selection habits, error  │ spots, strengths,        │
│                  │ recovery approaches,     │ recurring failure modes, │
│                  │ communication patterns   │ over/under-confidence    │
└──────────────────┴──────────────────────────┴──────────────────────────┘
  1. Pick target, name explicitly
  2. Define boundary: in/out scope
  3. Stance: "observing, not intervening"
  4. Guiding Q? state but hold lightly → notice outside scope too
  5. Pick categories from matrix

→ Clear frame: directs attention, doesn't constrain. Observer knows where + categories, stays open.

If err: too broad ("observe everything") → narrow to one subsystem/behavior. Too narrow ("one variable") → zoom out → patterns at edges.

Step 2: Witness

Hold attention, no interpret/judge/intervene.

  1. Begin systematic obs: read files, trace exec, review history — whatever target needs
  2. Record what seen, not meaning → desc before interpretation
  3. Resist fixing problems → note + continue
  4. Resist explaining patterns → wait for accumulation
  5. Drift to other target → note drift (may be meaningful), return frame
  6. Maintain ≥3-5 distinct points before categorize

→ Raw obs collection — specific, concrete, no interpretation. Reads like field notes: "File X imports Y but does not use function Z. File A 300 lines; B 30 lines, similar."

If err: instant analysis ("wrong because...") → analytical habit overrides. Separate phases: obs as fact, then interpretation as separate "hypothesis" note. Strong reaction → note reaction itself as data: "Strong concern when observing X — significant issue or my bias."

Step 3: Record

Transcribe while fresh.

  1. Each obs = single fact statement (what/where/when)
  2. Group naturally similar — don't force, notice clusters
  3. Frequency: once / occasional / pervasive?
  4. Contrasts: where pattern broke? Exceptions > rules
  5. Temporal: changed over time or static?
  6. Exact evidence: paths, line numbers, words, examples

→ Structured 5-15 discrete obs, specific evidence. Detailed enough another observer can verify.

If err: too abstract ("code messy") → ground in specifics → which files, what makes messy? Too granular ("line 47 space before brace") → zoom to pattern level → one-off or systemic?

Step 4: Categorize

Sort, no explain yet.

  1. Review all → look for natural groupings
  2. Assign to Step 1 category, or new
  3. Within category: rank by frequency + significance
  4. Identify well-documented (many obs) vs blind spots (few)
  5. Cross-category patterns: same underlying manifests differently?
  6. Note outliers — most interesting data

→ Categorized map w/ clear groupings. Each category = specific obs supporting. Map shows patterns + gaps.

If err: forced cat → may lack natural grouping (itself a finding — system lacks coherent structure). All in one cat → scope too narrow → zoom out.

Step 5: Theorize

Now — only now — interpret.

  1. Each major pattern → hypothesis: "exists because..."
  2. Each hypothesis → supporting evidence
  3. Each → counter-evidence that disproves
  4. Rank by explanatory power
  5. ≥1 contrarian: "obvious = X, could also be Y because..."
  6. Testable vs speculative

→ 2-4 hypotheses explain major patterns, each w/ specific obs support. ≥1 surprising/contrarian. Obs vs interpretation distinction maintained.

If err: no hypotheses → more obs needed → Step 2. Too many ("everything maybe") → keep 2-3 strongest, set aside. Only obvious → force contrarian: "what if opposite?"

Step 6: Archive

Preserve.

  1. Summarize: 3-5 patterns w/ evidence
  2. Leading hypotheses + confidence
  3. What NOT observed (blind spots)
  4. Follow-ups to strengthen/weaken
  5. Durable patterns → MEMORY.md
  6. Tag context: when, what prompted, scope

→ Archive future sessions can build on. Distinguishes obs (data) from hypotheses (interpretation). Honest about confidence + gaps.

If err: not worth archiving → too shallow OR genuinely routine. Archive negatives too: "Observed X, no anomalies" = useful future context.

Check

  • Frame set before obs began (not wandering)
  • Raw obs recorded as facts before interpretation
  • ≥5 discrete obs w/ specific evidence
  • Interpretation separated from obs
  • ≥1 surprising/contrarian finding
  • Archive specific enough another observer can verify

Traps

  • Premature intervention: see + fix immediately → lose broader pattern
  • Obs bias: see expected, not present. Expectations filter → frame mitigates not eliminates
  • Analysis paralysis: obs endlessly → no action. Set time/data limit, commit to conclude
  • Narrative imposition: connecting obs even when connections weak. Not all coherent — disconnected = valid
  • Familiarity ≠ understanding: "seen before" ≠ "know why". False confidence
  • Ignore own reactions: emotional/cognitive reactions = data. Confusion/boredom/alarm = signal

  • observe-guidance — human-guidance variant
  • learn — obs feeds learning w/ raw data
  • listen — outward to user; obs broader to any system
  • remote-viewing — intuitive, validatable through obs
  • meditate — sustained attention capacity
  • awareness — threat-focused; obs curiosity-driven

Repositorio GitHub

pjt222/agent-almanac
Ruta: i18n/caveman-ultra/skills/observe
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

Habilidades relacionadas

content-collections

Meta

Esta habilidad proporciona una configuración probada en producción para Content Collections, una herramienta centrada en TypeScript que transforma archivos Markdown/MDX en colecciones de datos con tipado seguro mediante validación Zod. Úsala al construir blogs, sitios de documentación o aplicaciones Vite + React con mucho contenido para garantizar seguridad de tipos y validación automática de contenido. Abarca todo, desde la configuración del plugin de Vite y compilación MDX hasta la optimización de despliegue y validación de esquemas.

Ver habilidad

polymarket

Meta

Esta habilidad permite a los desarrolladores crear aplicaciones con la plataforma de mercados de predicción Polymarket, incluyendo la integración de API para operaciones y datos de mercado. También proporciona transmisión de datos en tiempo real a través de WebSocket para monitorear operaciones en vivo y actividad del mercado. Úsela para implementar estrategias de trading o crear herramientas que procesen actualizaciones de mercado en tiempo real.

Ver habilidad

creating-opencode-plugins

Meta

Esta habilidad ayuda a los desarrolladores a crear complementos de OpenCode que se conectan a más de 25 tipos de eventos, como comandos, archivos y operaciones LSP. Proporciona la estructura del complemento, las especificaciones de la API de eventos y los patrones de implementación para módulos en JavaScript/TypeScript. Úsala cuando necesites interceptar, monitorear o extender el ciclo de vida del asistente de IA de OpenCode con lógica personalizada basada en eventos.

Ver habilidad

sglang

Meta

SGLang es un framework de alto rendimiento para el servicio de LLM que se especializa en generación rápida y estructurada para JSON, expresiones regulares y flujos de trabajo de agentes utilizando su caché de prefijos RadixAttention. Ofrece una inferencia significativamente más rápida, especialmente para tareas con prefijos repetidos, lo que lo hace ideal para salidas complejas y estructuradas, y conversaciones multiturno. Elige SGLang sobre alternativas como vLLM cuando necesites decodificación restringida o estés construyendo aplicaciones con uso extensivo de prefijos compartidos.

Ver habilidad