The Art of Agent Direction | ACFS Learning Hub

Goal

Master the art of directing AI agents with precision and intention.

Why Prompting Matters

The difference between a mediocre agent session and a brilliant one often comes down to how you direct the agent. This lesson dissects the patterns that make prompts effective, drawn from real-world workflows that consistently produce excellent results.

Intensity Calibration

Signal how much attention to allocate

Scope Control

Expand or contract the search space

Metacognition

Force self-verification and reflection

Context Anchoring

Ground behavior in stable references

Pattern 1: Intensity Calibration

AI models allocate "compute" based on perceived task importance. Stacked modifiers signal that this task deserves maximum attention:

"super carefully"→Elevates attention above baseline

"super careful, methodical, and critical"→Triple-stacking for maximum precision

"systematically and meticulously and intelligently"→Emphasizes both process and quality

bash

# Low intensity (default behavior)
"Check the code for bugs"
# High intensity (elevated attention)
"Do a super careful, methodical, and critical check
with fresh eyes to find any obvious bugs, problems,
errors, issues, silly mistakes, etc. and then
systematically and meticulously and intelligently
correct them."

Note

These aren't filler words. They're calibration signals that tell the model to allocate more reasoning depth to the task.

Pro Tip

Claude Code feature: The /effort command (/effort low/medium/high/max) controls how much thinking Claude Code allocates to a task. Use /effort max for tasks requiring maximum reasoning depth. While it's a tool-level feature in Claude Code, using intensity words like "think deeply" or "reason carefully" can help other agents/models allocate more attention to complex tasks as well.

Pattern 2: Scope Control

Models tend to take shortcuts. Explicit scope directives push against premature narrowing:

Breadth

"take ALL of that"
"Don't restrict yourself"
"cast a wider net"
"comprehensive and granular"

Depth

"go super deep"
"deeply investigate and understand"
"trace their functionality and execution flows"
"first-principle analysis"

bash

# Avoiding narrow focus
"Don't restrict yourself to the latest commits,
cast a wider net and go super deep!"
# Comprehensive coverage
"Take ALL of that and elaborate on it more,
then create a comprehensive and granular set..."
# Depth with breadth
"Randomly explore the code files in this project,
choosing code files to deeply investigate and understand
and trace their functionality and execution flows
through the related code files which they import
or which they are imported by."

Pattern 3: Forcing Self-Verification

Questions trigger metacognition—forcing the model to evaluate its own output before finalizing:

"Are you sure it makes sense?"

Basic sanity check

"Is it optimal?"

Pushes beyond 'good enough'

"Could we change anything to make the system work better for users?"

User-centric optimization

"Check over each bead super carefully"

Item-by-item review

bash

# The Plan Review Pattern
"Check over each bead super carefully—
are you sure it makes sense?
Is it optimal?
Could we change anything to make the system work better?
If so, revise the beads.
It's a lot easier and faster to operate in 'plan space'
before we start implementing these things!"

Pro Tip

Plan Space Principle: Revising plans is 10x cheaper than debugging implementations. Force verification at the planning stage.

Pattern 4: The Fresh Eyes Technique

Psychological reset techniques help agents approach code without prior assumptions or confirmation bias:

Explicit Reset

with "fresh eyes"

Signals to discard prior assumptions

Random Exploration

"randomly explore the code files"

Avoids tunnel vision on expected locations

Peer Framing

"reviewing code written by your fellow agents"

Creates psychological distance from own work

bash

# The Fresh Eyes Code Review
"I want you to carefully read over all of the new code
you just wrote and other existing code you just modified
with 'fresh eyes' looking super carefully for any obvious
bugs, errors, problems, issues, confusion, etc.
Carefully fix anything you uncover."
# Peer Review Framing
"Turn your attention to reviewing the code written by
your fellow agents and checking for any issues, bugs,
errors, problems, inefficiencies, security problems,
reliability issues, etc. and carefully diagnose their
underlying root causes using first-principle analysis."

Pattern 5: Temporal Awareness

Great prompts consider future contexts—the agent that will continue this work, the human who will review it, the "future self" who needs to understand it:

bash

# Self-Documenting Output
"Create a comprehensive set of beads with detailed comments
so that the whole thing is totally self-contained and
self-documenting (including relevant background,
reasoning/justification, considerations, etc.—
anything we'd want our 'future self' to know about
the goals and intentions and thought process and how it
serves the over-arching goals of the project)."

Future Self--Write as if explaining to someone with no context

Self-Contained--Output should work independently of current conversation

Over-Arching Goals--Connect current work to bigger picture

Pattern 6: Context Anchoring

Stable reference documents (like AGENTS.md) serve as behavioral anchors. Re-reading them is especially critical after context compaction.

bash

# The Post-Compaction Refresh
"Reread AGENTS.md so it's still fresh in your mind.
Use /effort max."

Warning

Why this matters after compaction:

1. Context decay: Rules lose salience as more content is added
2. Summarization loss: Compaction may miss nuances
3. Drift prevention: Periodic grounding prevents behavioral divergence
4. Fresh frame: Re-reading establishes correct operating context

bash

# Grounding Throughout Work
"Be sure to comply with ALL rules in AGENTS.md and
ensure that any code you write or revise conforms to
the best practice guides referenced in the AGENTS.md file."
# Making Rules Explicit
"You may NOT delete any file or directory unless I
explicitly give the exact command in this session."

Pattern 7: First Principles Analysis

Push for deep understanding over surface-level pattern matching:

bash

# Root Cause Emphasis
"Carefully diagnose their underlying root causes
using first-principle analysis and then fix or
revise them if necessary."
# Context Before Action
"Once you understand the purpose of the code in
the larger context of the workflows, I want you
to do a super careful, methodical check..."

Understand Before Fixing--Trace execution flows and dependencies first

Root Cause Over Symptom--Diagnose underlying issues, not surface manifestations

Larger Context--Understand how code fits into overall workflows

Putting It All Together

Here's a real prompt that combines multiple patterns:

markdown

"Reread AGENTS.md so it's still fresh in your mind.
Use /effort max.
I want you to sort of randomly explore the code files
in this project, choosing code files to deeply investigate
and understand and trace their functionality and execution
flows through the related code files which they import or
which they are imported by.
Once you understand the purpose of the code in the larger
context of the workflows, I want you to do a super careful,
methodical, and critical check with 'fresh eyes' to find
any obvious bugs, problems, errors, issues, silly mistakes,
etc. and then systematically and meticulously and
intelligently correct them.
Be sure to comply with ALL rules in AGENTS.md and ensure
that any code you write or revise conforms to the best
practice guides referenced in the AGENTS.md file."

Interactive Prompt Engineering Lab

Toggle techniques, watch prompts build layer by layer, and compare agent output quality.

Building prompt layer by layer (0/1)

Click "Start Building" to watch prompt layers stack up.
Toggle techniques above to customize which layers appear.

System Prompts

Set the agent's role, persona, and operating constraints before it sees the user task.

95%

180 tokens

Quick Reference

Pattern

Intensity

When

Tasks requiring maximum precision

Key Phrases

super carefully, methodical, use /effort max

Pattern

Scope Expansion

When

Avoiding narrow focus or shortcuts

Key Phrases

take ALL, cast wider net, comprehensive

Pattern

Self-Verification

When

Before implementing or finalizing

Key Phrases

are you sure?, is it optimal?, revise if needed

Pattern

Fresh Eyes

When

Code review, finding missed issues

Key Phrases

fresh eyes, fellow agents, randomly explore

Pattern

Temporal

When

Creating persistent artifacts

Key Phrases

future self, self-documenting, self-contained

Pattern

Anchoring

When

After compaction or drift risk

Key Phrases

reread AGENTS.md, comply with ALL rules

Pattern

First Principles

When

Debugging or understanding complex code

Key Phrases

root causes, first-principle, larger context

# Low intensity (default behavior)

"Check the code for bugs"

# High intensity (elevated attention)

"Do a super careful, methodical, and critical check

with fresh eyes to find any obvious bugs, problems,

errors, issues, silly mistakes, etc. and then

systematically and meticulously and intelligently

correct them."

# Avoiding narrow focus

"Don't restrict yourself to the latest commits,

cast a wider net and go super deep!"

# Comprehensive coverage

"Take ALL of that and elaborate on it more,

then create a comprehensive and granular set..."

# Depth with breadth

"Randomly explore the code files in this project,

choosing code files to deeply investigate and understand

and trace their functionality and execution flows

through the related code files which they import

or which they are imported by."

# The Plan Review Pattern

"Check over each bead super carefully—

are you sure it makes sense?

Is it optimal?

Could we change anything to make the system work better?

If so, revise the beads.

It's a lot easier and faster to operate in 'plan space'

before we start implementing these things!"

# The Fresh Eyes Code Review

"I want you to carefully read over all of the new code

you just wrote and other existing code you just modified

with 'fresh eyes' looking super carefully for any obvious

bugs, errors, problems, issues, confusion, etc.

Carefully fix anything you uncover."

# Peer Review Framing

"Turn your attention to reviewing the code written by

your fellow agents and checking for any issues, bugs,

errors, problems, inefficiencies, security problems,

reliability issues, etc. and carefully diagnose their

underlying root causes using first-principle analysis."

# Self-Documenting Output

"Create a comprehensive set of beads with detailed comments

so that the whole thing is totally self-contained and

self-documenting (including relevant background,

reasoning/justification, considerations, etc.—

anything we'd want our 'future self' to know about

the goals and intentions and thought process and how it

serves the over-arching goals of the project)."

# Grounding Throughout Work

"Be sure to comply with ALL rules in AGENTS.md and

ensure that any code you write or revise conforms to

the best practice guides referenced in the AGENTS.md file."

# Making Rules Explicit

"You may NOT delete any file or directory unless I

explicitly give the exact command in this session."

# Root Cause Emphasis

"Carefully diagnose their underlying root causes

using first-principle analysis and then fix or

revise them if necessary."

# Context Before Action

"Once you understand the purpose of the code in

the larger context of the workflows, I want you

to do a super careful, methodical check..."

"Reread AGENTS.md so it's still fresh in your mind.

Use /effort max.

I want you to sort of randomly explore the code files

in this project, choosing code files to deeply investigate

and understand and trace their functionality and execution

flows through the related code files which they import or

which they are imported by.

Once you understand the purpose of the code in the larger

context of the workflows, I want you to do a super careful,

methodical, and critical check with 'fresh eyes' to find

any obvious bugs, problems, errors, issues, silly mistakes,

etc. and then systematically and meticulously and

intelligently correct them.

Be sure to comply with ALL rules in AGENTS.md and ensure

that any code you write or revise conforms to the best

practice guides referenced in the AGENTS.md file."

New to ACFS?

Why Prompting Matters

Intensity Calibration

Scope Control

Metacognition

Context Anchoring

Pattern 1: Intensity Calibration

Pattern 2: Scope Control

Breadth

Depth

Pattern 3: Forcing Self-Verification

Pattern 4: The Fresh Eyes Technique

Explicit Reset

Random Exploration

Peer Framing

Pattern 5: Temporal Awareness

Pattern 6: Context Anchoring

Pattern 7: First Principles Analysis

Putting It All Together

Interactive Prompt Engineering Lab

Quick Reference

Ready to level up?

Why Prompting Matters

Intensity Calibration

Scope Control

Metacognition

Context Anchoring

Pattern 1: Intensity Calibration

Pattern 2: Scope Control

Breadth

Depth

Pattern 3: Forcing Self-Verification

Pattern 4: The Fresh Eyes Technique

Explicit Reset

Random Exploration

Peer Framing

Pattern 5: Temporal Awareness

Pattern 6: Context Anchoring

Pattern 7: First Principles Analysis

Putting It All Together

Interactive Prompt Engineering Lab

Quick Reference