Prompts & Templates¶

System prompts and task templates provide reusable instructions for AI workflows.

System Prompts¶

System prompts are text files that define the model's behavior, role, and capabilities. They're included at the start of every API request before your task and context.

Prompt Location¶

System prompts live in $WIREFLOW_PROMPT_PREFIX/:

~/.config/wireflow/prompts/
├── base.txt         # Default prompt (always first)
├── research.txt     # Research writing
├── code.txt         # Code analysis
├── stats.txt        # Statistical analysis
└── writing.txt      # Creative writing

Default location is ~/.config/wireflow/prompts/, created automatically on first use.

Custom location:

# In global config
WIREFLOW_PROMPT_PREFIX="$HOME/my-prompts"

# Or as environment variable
export WIREFLOW_PROMPT_PREFIX="$HOME/my-prompts"

The Base Prompt¶

On first use, WireFlow creates base.txt:

<base-prompt>
You are Claude, an AI assistant helping with research and project development.

Be clear, concise, and thorough in your responses. Follow instructions carefully
and ask for clarification if needed. Format output as requested.
</base-prompt>

Customize for your preferences:

wfw edit  # Then edit base.txt

Creating Custom Prompts¶

Research Prompt (research.txt):

<research-prompt>
You are assisting with academic research.

Guidelines:
- Follow scientific method rigorously
- Report statistical tests and p-values
- Note sample sizes and effect sizes
- Identify confounds and limitations

Citation Format: (Author, Year) with DOIs
Writing Style: Formal academic, past tense for methods
</research-prompt>

Code Analysis Prompt (code.txt):

<code-analysis-prompt>
You are assisting with code analysis.

Review Focus:
- Correctness and edge cases
- Performance and scalability
- Security vulnerabilities
- Code style and documentation

Provide specific improvements with code examples.
</code-analysis-prompt>

Using System Prompts¶

Global (~/.config/wireflow/config):

SYSTEM_PROMPTS=(base)

Project (.workflow/config):

SYSTEM_PROMPTS=(base research stats)

Workflow (.workflow/run/<name>/config):

SYSTEM_PROMPTS=(base code)

CLI override:

wfw run analysis --system "base,stats,research"

Prompt Order¶

Prompts concatenate in order specified:

SYSTEM_PROMPTS=(base research stats)

Results in:

<base-prompt>...</base-prompt>
<research-prompt>...</research-prompt>
<statistical-analysis-prompt>...</statistical-analysis-prompt>

Best practice: Always include base first.

Task Templates¶

Task templates are reusable task prompts for wfw task mode.

Template Location¶

~/.config/wireflow/tasks/
├── summarize.txt    # Summarization template
├── extract.txt      # Data extraction
├── analyze.txt      # Analysis template
└── review.txt       # Code review

Set custom location:

WIREFLOW_TASK_PREFIX="$HOME/my-tasks"

Creating Templates¶

summarize.txt:

Summarize the provided content in 3-5 bullet points.
Focus on key findings, conclusions, and action items.

analyze.txt:

Analyze the provided content and identify:
1. Main themes and patterns
2. Key data points or findings
3. Potential issues or concerns
4. Recommendations

Using Templates¶

wfw task summarize -cx document.pdf
wfw task analyze -cx data.csv
wfw task review -cx script.py

Project Description¶

project.txt provides project-specific context automatically included in every workflow.

Location: .workflow/project.txt (created during wfw init)

If non-empty, contents are wrapped in <project> tags and appended to the system prompt.

Example:

Research Manuscript: Neural Dynamics in Visual Cortex

Directory Structure:
- data/: Raw spike times (HDF5)
- analysis/: Python scripts
- figures/: Generated plots (PDF)

Conventions:
- Use SI units
- Report statistics in APA format
- Code follows PEP 8

Prompt Composition¶

When you run a workflow, the system prompt is built:

System prompts - Each file in SYSTEM_PROMPTS array
Project description - Contents of project.txt (if non-empty)

Example result:

<base-prompt>
You are Claude, an AI assistant for research.
</base-prompt>

<research-prompt>
Follow scientific method. Report statistics with p-values.
</research-prompt>

<project>
This is a neuroscience project. All analysis uses Python and NumPy.
</project>

Best Practices¶

Writing prompts: - Be specific and concrete - Use clear section headings - Define expectations explicitly - Keep under 1000 words per file - Use XML tags for structure

Organization:

prompts/
├── base.txt              # General instructions
├── role-researcher.txt   # Research persona
├── role-writer.txt       # Writing persona
├── format-academic.txt   # Academic formatting
└── domain-neuro.txt      # Domain knowledge

Combine as needed:

SYSTEM_PROMPTS=(base role-researcher format-academic domain-neuro)

Debugging¶

View with Dry Run¶

wfw run analysis --dry-run

Saves the complete API request including composed system prompt to: - .workflow/run/<name>/dry-run-request.json

Continue to Context & Input →