Skip to content

Docs/agent prompt principles#15

Merged
mpawliszyn merged 2 commits intomainfrom
docs/agent-prompt-principles
Feb 27, 2026
Merged

Docs/agent prompt principles#15
mpawliszyn merged 2 commits intomainfrom
docs/agent-prompt-principles

Conversation

@mpawliszyn
Copy link
Copy Markdown
Collaborator

No description provided.

mpawliszyn and others added 2 commits February 26, 2026 16:16
12 principles for writing effective agent prompts, derived from research
on multi-agent orchestration, cognitive science, LLM behavioral studies,
and production agent tools. Covers:

- Agent-as-tool over agent-as-peer (Anthropic recommendation)
- Format constraints as behavioral control (Aider: 3x improvement)
- Negative constraints before positive instructions (RPI pattern)
- Named anti-rationalization tables (Superpowers pattern)
- Mechanical verification over self-assessment
- Cognitive load limits for tree structure (Miller/Cowan research)
- Unified diff format for agent consumption (Diff-XYZ benchmark)
- Fresh start design for long sessions (39% multi-turn degradation)
- Supervisor mode for parallel agents
- The reviewer as protagonist (anti-automation-bias)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Comprehensive research summary covering the evidence base behind Fowlcon
design decisions. Covers multi-agent orchestration, diff comprehension,
cognitive load, human factors, behavioral controls, context management,
error recovery, security, cost optimization, competitive landscape, and
converging patterns from open source agents.

All sources are public (papers, official docs, open source repos).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@mpawliszyn mpawliszyn merged commit 3ebc25c into main Feb 27, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant