Context Prime
Load task-specific context dynamically based on task type.
Arguments
$ARGUMENTS: Task type (bug, feature, review, chore, research) or custom description
Instructions
You are loading task-specific context to prime the agent for focused work.
Step 1: Determine Task Type
Parse $ARGUMENTS to determine task type:
| Type | Context Focus |
|---|---|
| bug | Recent commits, test files, error patterns |
| feature | Architecture, related modules, API patterns |
| review | Style guide, test patterns, changed files |
| chore | Project structure, tooling, configs |
| research | Documentation, external resources, examples |
| (custom) | Infer appropriate context from description |
Step 2: Execute Discovery
For all types - Run:
git status
git ls-files | head -50
Type-specific discovery:
bug:
git log --oneline -10
git diff HEAD~5 --stat
feature:
ls -la src/
ls -la lib/
review:
git diff --stat
git log --oneline -5
chore:
ls -la
cat package.json | head -20 # or equivalent manifest
research:
ls -la docs/ 2>/dev/null || echo "No docs directory"
ls -la ai_docs/ 2>/dev/null || echo "No ai_docs directory"
Step 3: Load Essential Files
For all types - Read:
- README.md (if exists)
Type-specific reads:
bug:
- Recent test files related to area
- Configuration files
feature:
- Architecture documentation
- Similar feature implementations
review:
- Style guide or linting config
- Test patterns
chore:
- Tooling documentation
- Build/deploy configs
research:
- Existing documentation
- Reference materials
Step 4: Report Understanding
Output
Summarize loaded context:
## Context Primed: [Type]
**Task Type:** [bug/feature/review/chore/research]
**Files Loaded:** [count]
### Project State
- Branch: [current branch]
- Status: [clean/dirty]
- Recent changes: [summary]
### Loaded Context
- [File 1]: [brief description]
- [File 2]: [brief description]
### Ready For
[What this priming prepares you for]
### Context Efficiency
This priming loaded ~[X] tokens of task-specific context.
Remaining capacity: [estimate]
Notes
- Priming loads fresh, relevant context for the task at hand
- Prefer priming over static CLAUDE.md for task-specific work
- See @context-priming-patterns.md for pattern details
- See @rd-framework.md for reduce and delegate strategies