Skip to main content
context-engineering Context window monitoring and budget management. Keeps orchestrator at 15-30% context usage while subagents get full 200k tokens. Provides warnings at thresholds, context-aware summarization triggers, and wave-level budget planning.
Ir a la instalación Skills Marketplace Descubre y explora habilidades de IA creadas por la comunidad.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Mostrar detalles del prompt Un comando directo omite el prompt de revisión. Revisa el origen antes de ejecutarlo.
npx skills add https://github.com/a5c-ai/babysitter --skill context-engineeringEl comando permanece en una sola línea. Desplázate horizontalmente para revisarlo antes de copiarlo.
¿Prefieres una copia local? Descarga los archivos que SkillsMP tiene disponibles ahora.
Descargar Zip Descargando... Más de este repositorio Reference for querying the Atlas knowledge graph through its MCP tools — the SECONDARY enrichment/comparison layer that adds best-practice context to systems you have ALREADY scanned from your real sources (`az`, repos, dirs). Use when you need to look up nodes, edges, kinds, clusters, stats, or wiki pages in Atlas to compare against your real inventory. (atlas graph, query atlas, atlas mcp, search the graph, graph neighbors, atlas record, atlas kinds, enrichment layer)
Atlas turns your STATED NEED into a real systems atlas by SCANNING your actual sources (Azure via `az`, git repos, local dirs) and process/data mining them, THEN enriching against the Atlas knowledge graph. Use this skill when asked to inventory/map your real systems, scan your cloud + repos + directories, mine the real processes or data they contain, or collect their real constraints/gotchas. (atlas, scan my systems, inventory our azure account, map my repos, real systems atlas, process mining, data mining, collect nuances, system discovery)
assimilate-popular-workflows This skill should be used when the user asks to "find skills in the wild", "assimilate popular workflows", "discover SKILL.md files in repos", "research external skills", "find workflow patterns", "survey the skill landscape", "what skills exist out there", or wants to investigate public repositories for extractable processes, babysitter plugins, and reusable procedural insights. Searches GitHub for SKILL.md files, classifies repos by archetype, and maintains structured research under docs/reference-repos/.
Explorador de archivos
2 archivos Ocupaciones relacionadas SOC
Basado en la clasificación ocupacional SOC
name context-engineering description Context window monitoring and budget management. Keeps orchestrator at 15-30% context usage while subagents get full 200k tokens. Provides warnings at thresholds, context-aware summarization triggers, and wave-level budget planning. allowed-tools Read Bash(*) metadata {"author":"babysitter-sdk","version":"1.0.0","category":"gsd-core","backlog-id":"SK-GSD-003"} graph {"domains":["domain:software-engineering"],"skillAreas":["skill-area:agentic-loops","skill-area:orchestration-loop"],"workflows":["workflow:feature-development"],"topics":["topic:developer-experience"],"roles":["role:tech-lead","role:backend-engineer"]}
Subagents : Get full 200k tokens of fresh context per spawn.Context budget : Plan how much context each wave of execution will consume.
This skill provides:
Context window usage estimation for the current session
Warning injection at configurable thresholds (70%, 85%, 95%)
Orchestrator budget enforcement
Subagent context allocation recommendations
Context-aware summarization triggers
Stale context detection and pruning suggestions
Wave-level context budget planning
Capabilities
1. Context Usage Estimation Estimate current context window usage based on conversation history size:
Tokens used: ~45,000 / 200,000
Usage: 22.5%
Status: HEALTHY
Next threshold: 70% (warning) at ~140,000 tokens
Character count / 4 (rough approximation)
Tool output tracking (each tool call adds to context)
File read accumulation tracking
2. Threshold Warnings Inject warnings at configurable thresholds:
[CONTEXT 70%] Warning: Context window at 70%. Consider summarizing completed work.
[CONTEXT 85%] Critical: Context window at 85%. Spawn new subagent for remaining work.
[CONTEXT 95%] Emergency: Context window at 95%. Wrap up immediately. Write state and exit.
70% : Suggest summarizing completed work, pruning stale context85% : Strongly recommend spawning new subagent with fresh context95% : Emergency wrap-up: write STATE.md, commit, create continue-here.md
3. Orchestrator Budget Enforcement Monitor orchestrator-specific budget:
Target orchestrator usage: 15-30%
Current orchestrator usage: 18%
Remaining budget: 12% (~24,000 tokens)
Budget allocation:
- Phase context loading: 5% (PROJECT, ROADMAP, STATE)
- Agent spawn overhead: 3% per agent
- Result processing: 2% per agent result
- State updates: 1%
4. Subagent Context Allocation Recommend context allocation for subagent spawns:
Agent: gsd-executor
Available context: 200,000 tokens (fresh)
Recommended loading:
- Plan file: ~2,000 tokens
- Relevant source files: ~15,000 tokens
- Project context: ~3,000 tokens
- Remaining for execution: ~180,000 tokens
5. Context-Aware Summarization Trigger summarization when context is filling:
Summarization triggers:
- Tool output > 10,000 characters: summarize before continuing
- File read > 5,000 lines: extract relevant sections only
- Agent result > 20,000 characters: summarize key outcomes
Summarization strategies:
Completed work : Replace detailed execution logs with summaryFile contents : Replace full file reads with relevant excerptsAgent results : Extract key outcomes, discard detailed reasoning
6. Stale Context Detection Identify context that is no longer relevant:
Stale context candidates:
- File contents read 10+ interactions ago
- Agent results from completed (not current) phases
- Tool outputs that were informational only
- Research documents already synthesized into plans
7. Wave-Level Budget Planning Plan context budget across execution waves:
Wave 1 (3 parallel agents):
Spawn cost: 3 * 3% = 9%
Result processing: 3 * 2% = 6%
Wave total: 15%
Wave 2 (2 parallel agents):
Spawn cost: 2 * 3% = 6%
Result processing: 2 * 2% = 4%
Wave total: 10%
Total orchestrator budget needed: 25%
Target: 30% -> Sufficient with 5% margin
Tool Use Instructions
Checking Context Usage
Use Bash to estimate current session token count if available
Track cumulative tool output sizes during the session
Calculate percentage against 200,000 token window
Return usage report with threshold proximity
Injecting Warnings
Compare current usage against configured thresholds
If threshold exceeded, format appropriate warning message
Include recommended action based on threshold level
For 95%: include emergency state-save instructions
Planning Wave Budgets
Use Read to load plan files for the phase
Count agents needed per wave
Estimate per-agent spawn and result cost
Sum wave costs and compare to orchestrator budget target
Recommend wave splitting if budget exceeded
Process Integration
execute-phase.js - Monitor context during multi-wave execution, trigger summarization between wavesiterative-convergence.js - Track context across convergence iterations, spawn fresh agents when context fillsnew-project.js - Budget context for parallel research agents (4 spawns + synthesis)
Output Format {
"operation" : "check|warn|plan|summarize" ,
"status" : "healthy|warning|critical|emergency" ,
"usage" : {
"estimatedTokens" : 45000 ,
"maxTokens" : 200000 ,
"percentage" : 22.5 ,
"nextThreshold" : 70
} ,
"recommendation" : "Continue normally|Summarize completed work|Spawn new agent|Emergency wrap-up" ,
"waveBudget" : {
"totalWaves" : 2 ,
"estimatedOrchestratorUsage" : 25 ,
"withinBudget" : true }
}
Configuration Setting Default Description contextWarningThreshold70Warning threshold percentage contextCriticalThreshold85Critical threshold percentage contextEmergencyThreshold95Emergency threshold percentage orchestratorBudgetTarget30Target max orchestrator context % agentSpawnCost3Estimated % per agent spawn overhead agentResultCost2Estimated % per agent result processing autoSummarizetrueAuto-trigger summarization at thresholds
Error Handling Error Cause Resolution Token estimation inaccurateApproximation drift Use conservative estimates (overestimate usage) Budget exceeded mid-waveUnderestimated agent costs Defer remaining agents to next session with continue-here.md Emergency threshold hitOrchestrator doing too much work inline Immediately write state, commit, create handoff document Stale context false positiveContext still needed Maintain a "pinned context" list that is never pruned
Constraints
Token estimates are approximations (character/4 heuristic); always err on the side of caution
Never discard context that has not been persisted to disk (STATE.md, summaries, etc.)
Orchestrator should never exceed 30% context usage
Emergency wrap-up at 95% is non-negotiable; quality degrades severely above this
Wave budget planning must account for worst-case agent result sizes
Context monitoring is advisory; it cannot forcibly stop execution
