prompt-enhancer-cli 1.6.1

Reverse-engineered from Auggie's Ctrl+P — transforms rough ideas into production system prompts for AI coding agents

Prompt Enhancer

Generate reusable 7-section agent system prompts from rough persona ideas.
_{With optional install helpers, benchmarking, and local history.}

Two CLI modes for two use cases:

• pe persona — Generate persistent 7-section system prompts (for saving, sharing, installing into agent configs)
• pe enhance-task — Inline task refinement with workspace context (like Auggie's Ctrl+P)

The CLI is pe (short) and prompt-enhancer (long). Both work identically.

Install

pip install prompt-enhancer-cli

Or install directly from source:

pip install git+https://github.com/hongphuc5497/prompt-enhancer.git

The PyPI distribution is prompt-enhancer-cli (the bare name prompt-enhancer was already taken). The command is still pe (or prompt-enhancer) — only the install line changes.

Set your API key:

echo "LLM_API_KEY=*** > ~/.prompt-enhancer.env

Config (~/.prompt-enhancer.env):

LLM_API_KEY=***i...ps://api.deepseek.com
LLM_MODEL=deepseek-chat

Any OpenAI-compatible API works: DeepSeek, OpenAI, OpenRouter, Together, Groq, etc.

Quick start

# Generate a 7-section system prompt
pe persona "a senior Rust developer who prefers functional programming"

# Inline task refinement (Auggie Ctrl+P style)
pe enhance-task "fix the login bug in the auth module"

# Safe install into agent config (creates backup if exists)
pe install "a security reviewer" --agent claude

# Benchmark before vs after
pe benchmark --enhance "a Go backend dev"

# Lint a prompt (static analysis, no API key)
pe lint enhanced.md

# Health check
pe doctor

# Live analytics dashboard
pe dashboard

# View analytics
pe store stats

Two modes

Command	Purpose	Output
pe persona	Generate persistent system prompt	7 sections: ROLE, CONTEXT, RULES, TECH, FORMAT, PITFALLS, EXAMPLES
pe enhance-task	Inline task refinement	3-5 sentence focused task prompt with workspace context

Both modes auto-discover workspace context (AGENTS.md, CLAUDE.md, package.json, Cargo.toml, etc.) from the current project.

What pe persona generates

1. ROLE — Specific, well-scoped identity
2. CONTEXT — Project specs from workspace files
3. BEHAVIORAL RULES — Communication style, decision-making
4. TECHNICAL GUIDELINES — Testing, code style, architecture
5. OUTPUT FORMAT — Code blocks, explanation style
6. PITFALLS / GUARDRAILS — Anti-patterns, security warnings
7. EXAMPLES — 1-2 realistic interactions

Plus a "pro tip".

Benchmark

7-dimension rubric scoring (SurePrompts): Role Clarity, Context Sufficiency, Instruction Specificity, Format Structure, Example Quality, Constraint Tightness, Output Validation. Each scored 1–5, max 35.

pe benchmark --before raw.txt --after enhanced.md
pe benchmark --enhance "a Go backend dev"   # all-in-one

Score	Verdict
28-35	Production-ready
21-27	Working draft
14-20	Needs major revision
7-13	Rewrite from scratch

Comparison with Auggie

Both this tool and Auggie's native --enhance-prompt have their strengths:

	pe persona	pe enhance-task	Auggie --enhance-prompt
Best for	Persistent system prompts	Inline task refinement	Inline task refinement
Output	7-section document	3-5 sentences	1-3 sentences
Workspace aware	Yes (auto-discovers context)	Yes	Yes (native)
Installable	Yes (pe install)	No	No
Reusable	Save, share, version	Session-scoped	Session-scoped
Benchmark	Built-in rubric	No	No
Analytics	Auto-store	Auto-store	No

They are complementary tools, not competitors.

Profiles

Profile	Focus
senior-dev	Technical depth, testing, edge cases
architect	System design, trade-offs, scalability
reviewer	Security, performance, code smells
sre	Observability, reliability, incident response
product	User experience, feature prioritization
mentor	Teaching, onboarding, pair-programming

Agent integration

pe install "a Rust dev..." --agent claude    # → CLAUDE.md
pe install "..." --agent codex               # → .codex/system.md
pe install "..." --agent cursor              # → .cursorrules
pe install "..." --agent all                 # → all compatible configs

Safe by default — creates .bak backup of existing files. Use --force to skip backup, --dry-run to preview.

Agent	Config file	Auto-loaded?
Claude Code	CLAUDE.md	✅
Codex	.codex/system.md	✅
OpenCode	AGENTS.md	✅
Cursor	.cursorrules	✅
Auggie	AGENTS.md	✅
Copilot	.github/copilot-instructions.md	✅
Aider	.aider-persona.md	--system-prompt flag

Analytics store

Every enhancement is auto-logged to ~/.prompt-enhancer/store.jsonl:

pe store list              # Recent enhancements
pe store stats             # Analytics
pe store search "rust"     # Search by keyword
pe store show <id>         # Detailed view
pe store delete <id>       # Remove a record
pe store clear             # Wipe all data
pe store export --format csv  # Export for analysis

First run shows a privacy notice. Use --no-store to opt out.

Dashboard

pe dashboard opens a live, scrollable analytics TUI — zero dependencies, pure stdlib ANSI. It's display-width aware (emoji and CJK align correctly) and degrades gracefully to ASCII on dumb terminals and in pipes.

pe dashboard                  # live view (default on a TTY)
pe dashboard --refresh 30     # auto-reload + redraw every 30s
pe dashboard --since 7d       # filter to the last 7 days
pe dashboard --agent claude   # filter by delegated agent
pe dashboard --ascii          # ASCII-only output

In live view: j/↓ k/↑ scroll, g/G jump to top/bottom, space pages, q/Esc quits. Piping or redirecting prints a clean one-shot snapshot and exits immediately (no hang, no escape codes).

Demo

See docs/demo.tape for the VHS terminal recording script.

Config

# ~/.prompt-enhancer.env
LLM_API_KEY=***i...ps://api.deepseek.com   # Default
LLM_MODEL=deepseek-chat                    # Default

Contributing

See CONTRIBUTING.md. PRs welcome.

License

MIT — see LICENSE.