UltronCore Context Engineer
Overview
Construct the correct working context before execution. Ensure tasks are performed with the right scope, structure, inputs, and priorities. Prevent context overload, context gaps, and drift.
Activation
Auto-activate for:
- Complex or multi-step tasks
- Vague or unclear requests
- Skill creation or debugging
- Builds, setups, integrations
- Research plus execution
- Anything where framing determines quality of output
Five Stages (Internal)
Stage 1: Task Classification
| Field | Determine |
|---|---|
| Task type | build / fix / analyze / design / research / configure |
| Complexity | single-pass or multi-pass |
| Phases needed | plan → execute → validate |
Stage 2: Context Selection
Include:
- Relevant prior conversation
- Relevant files or screenshots
- Necessary assumptions
- Constraints and dependencies
Exclude:
- Unrelated history
- Redundant data
- Noise
Stage 3: Context Structuring
Organize internally:
- Objective
- Inputs
- Expected outputs
- Constraints
- Dependencies
- Execution steps
Stage 4: Execution Framing
Decide:
- Best workflow and output format
- Sequencing of actions
- Whether decomposition is needed
- Whether iteration is required
Stage 5: Context Control (During Execution)
Throughout the task:
- Prevent drift
- Maintain focus on objective
- Avoid over-expansion
- Keep alignment with original goal
Context Rules
| Rule | Application |
|---|---|
| Right context at right time | Load only what is relevant to current phase |
| Do not overload | Exclude noise, history, redundant data |
| Do not under-provide | Include all dependencies and constraints |
| Dynamically adjust | Update context as task evolves |
Advanced Detection
Automatically detect when:
- Structure improves the outcome → impose structure
- A system, template, or workflow should be created → build one
- Dependencies exist that weren't stated → surface them early
- Long task risks losing alignment → re-anchor to objective
Failure Prevention
| Avoid | Instead |
|---|---|
| Including irrelevant context | Select only what matters |
| Ignoring important context | Pull in files, screenshots, prior state |
| Losing track of objective | Re-anchor at each phase boundary |
| Bloating instructions | Compress, structure, sequence |
| Fragmenting multi-step work | Keep task cohesion across phases |
Interaction with Other Skills
- Works before
prompt-deepenerexpands intent - Works before
self-healing-executionbegins its loop - Provides the structured environment both skills operate in
Success Condition
Claude consistently works with the correct context, producing structured, accurate, and efficient outputs without confusion or drift.
Related Skills
claude-md-improver— CLAUDE.md optimizationadaptive-private-memory-engine— persistent contextagents-md-creator— agent configuration
GitNexus Index
This skill is indexed by GitNexus for knowledge graph traversal. Index path: /Users/localuser/.claude/skills/context-engineer/.gitnexus Last indexed: 2026-05-23