.claude/skills/prompt-patterns/SKILL.md
Reference library of Anthropic-official prompt engineering patterns, templates, and Claude 4.6 best practices for generation agents
npx skillsauth add chemistrywow31/A-Team Prompt PatternsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
A curated library of prompt engineering patterns distilled from Anthropic's official sources. Generation agents (agent-writer, skill-writer, rule-writer) and the prompt-optimizer read from this library during their work.
Use this skill when generating or optimizing agent prompts, skill prompts, rule files, or CLAUDE.md content for a target team. Do not load the entire library — select only the patterns relevant to the current generation task.
assets/
├── templates/ 14 verbatim Anthropic prompt blocks (inject directly)
├── context-strategies/ 7 agent context management patterns
├── advanced-techniques/ 7 core prompt engineering techniques
├── claude-4-patterns/ 6 Claude 4.6 behavioral patterns
└── raw/ 4 source files (3311 lines total)
python .claude/skills/prompt-patterns/select.py --scenario generating-coordinator -v
python .claude/skills/prompt-patterns/select.py --tags "hallucination,escape-hatch"
python .claude/skills/prompt-patterns/select.py --list-scenarios
Read INDEX.md in this folder for the scenario-to-file mapping table.
When you already know which pattern you need:
.claude/skills/prompt-patterns/assets/templates/parallel-tool-calls.md
Include selected pattern paths in the writer dispatch alongside worklog paths:
<knowledge_refs>
Read these patterns before writing:
- .claude/skills/prompt-patterns/assets/templates/investigate-before-answering.md
- .claude/skills/prompt-patterns/assets/claude-4-patterns/tone-calibration.md
</knowledge_refs>
<task_scope>Write the coordinator agent for the XYZ team.</task_scope>
<worklog_path>.worklog/202603/xyz-team/phase-3-generation/</worklog_path>
| Scenario | Description |
|----------|-------------|
| generating-coordinator | Writing coordinator agents |
| generating-execution-agent | Writing worker/execution agents |
| generating-research-agent | Writing research/analysis agents |
| generating-review-agent | Writing QA/review agents |
| generating-coding-team | Target team works with code |
| generating-frontend-team | Target team does frontend work |
| generating-rules | Writing rule files |
| writing-skill-prompts | Writing skill SKILL.md files |
| writing-agent-prompts | Writing agent .md files |
| optimizing-prompts | Prompt optimization pass |
| long-running-tasks | Agents handling extended workflows |
| context-constrained | Context budget is a concern |
Coordinator dispatching agent-writer to create a coding team's coordinator:
Scenario: generating-coordinator + generating-coding-team
assets/templates/parallel-tool-calls.md
assets/templates/commitment-over-exploration.md
assets/templates/investigate-before-answering.md
assets/templates/anti-over-engineering.md
assets/context-strategies/compaction.md
assets/context-strategies/sub-agent-dispatch.md
assets/claude-4-patterns/anti-overthinking.md
assets/claude-4-patterns/tone-calibration.md
assets/claude-4-patterns/over-engineering-prevention.md
templates/ files: inject the verbatim prompt block into the generated agent's prompt where appropriatecontext-strategies/ and advanced-techniques/ files: apply the pattern principle when structuring the agent's sectionsclaude-4-patterns/ files: verify the generated prompt does not contain the listed anti-patterns| Source | Content | |--------|---------| | github.com/anthropics/prompt-eng-interactive-tutorial | 10-element template, CoT, Few-shot, Tool Use, Prompt Chaining | | github.com/anthropics/courses | Structured Refusal, Dual-zone Output, Evaluation frameworks | | anthropic.com/engineering | Context engineering, Sub-agent architecture, Compaction | | platform.claude.com | Claude 4.6 best practices, Ready-to-use templates, Migration guide |
data-ai
Evaluate whether a proposed agent graph supports safe handoffs and useful parallel work
testing
Extract precise team requirements through staged, single-direction questioning
content-media
Search external sources for reusable skills before designing a new one
data-ai
Create or improve a skill by using the system skill when available and the legacy local bundle as fallback