maestria-co/context-loader

Skill: Context Loader

Purpose

Efficiently load the right .context/ files for the current task without reading everything. Context windows are finite — loading irrelevant files wastes capacity that should go to actual work.

This skill defines the loading order and selection strategy for different task types.

---

Context Discovery Order

Always read in this order. Stop as soon as you have enough context for the task.

Level 1 — Always Read (every task)

1. CLAUDE.md or .github/copilot-instructions.md — big picture, key commands
2. .context/overview.md — tech stack, architecture, current state

Level 2 — Read Based on Task Type

| Task type | Read these .context/ files | | ---------------------- | ---------------------------------------------------------------------- | | Feature implementation | domains/ (relevant entity), architecture.md, standards.md | | Bug fix | domains/ (affected area), testing.md, standards.md | | Refactoring | architecture.md, standards.md, standards.md | | Test writing | testing.md, testing.md, standards.md | | Architecture decision | architecture.md, decisions/index.md, all relevant domains/ files | | New domain work | domains/entities.md, domains/glossary.md, architecture.md | | CI/CD changes | workflows/ci-cd.md, workflows/branching.md | | UI/styling work | styling.md (if frontend project), standards.md |

Level 3 — Read When Resuming

1. .context/retrospectives/ — last few entries for recent lessons
2. .context/tasks/[TASK-ID]/plan.md — if resuming a specific task
3. git log --oneline -10 — recent commit history

---

Reading Strategies

Scan, Don't Read Everything

For large .context/ directories:

1. List files first: ls .context/domains/
2. Read file headers (first 20 lines) to determine relevance
3. Read fully only the files that match your task

Search Specifically

If you know what you're looking for:

• grep for entity names, pattern names, or error codes across .context/
• Don't read entire files hoping to stumble on relevant content

Cache What You Found

After loading context, note the key facts in your working memory:

• Tech stack and version constraints
• Naming conventions for the area you're working in
• Relevant business rules or domain constraints
• Applicable ADRs

---

When to Create New Context

During or after a task, create new .context/ files when:

| Situation | Action | | -------------------------------------- | ----------------------------------------------------------- | | Working in an undocumented domain area | Create domains/[area-name].md | | Discovering a non-obvious pattern | Add to architecture.md | | Finding inconsistent error handling | Document the correct pattern in standards.md | | Making an architecture decision | Create decisions/ADR-NNN-title.md via designing-systems | | Learning a lesson worth preserving | Add entry to retrospectives/ |

---

Constraints

• Do not read all .context/ files at the start of every task — be selective
• Do not spend more than 2 minutes on context loading — if you can't find what you need, proceed and note the gap
• Do not modify .context/ files during loading — that's context-maintenance
• Prefer targeted searches over full-file reads for large context directories