CLI Reference

CLI Reference — every command, subcommand and flag

Complete command-line reference for Wild_Root_Prompt: the seven subcommands, all generation flags, technique selection syntax, and copy-paste recipes for common jobs.

Every command, subcommand and flag. Run with no arguments to get the interactive menu instead:

python3 prompt_expert_enhance.py

---

Subcommands

Command	What it does
generate	One model, one manifest or enhanced prompt
parallel	Two models at once, split-screen streaming
full	parallel + expert synthesis, in one command
synthesis	Merge two manifest files you already have
templates	List or show the built-in starter templates
memory	View or clear session memory
web	Launch the local Web UI

---

generate

python3 prompt_expert_enhance.py generate "TASK" [options]

Option	Description
TASK	Task description. Optional if --template is used.
--topic TOPIC	Focus topic. Repeatable.
--model NAME	Ollama model to use.
--output FILE	Write here instead of an auto-named file in outputs/.

parallel

python3 prompt_expert_enhance.py parallel "TASK" --model-a llama3 --model-b qwen2.5:7b

Runs both models concurrently with live tokens in two columns. Best results come from pairing models with different reasoning styles — one systematic, one creative.

Option	Description
--model-a NAME	First model
--model-b NAME	Second model
--topic TOPIC	Focus topic. Repeatable.

full

python3 prompt_expert_enhance.py full "TASK" --techniques "1-30"

Parallel generation followed immediately by expert synthesis. The single most complete pipeline.

Option	Description
--model-a NAME / --model-b NAME	The two generation models
--synthesis-model NAME	Model that merges the two outputs

synthesis

python3 prompt_expert_enhance.py synthesis "ORIGINAL TASK" file_a.md file_b.md

Merges two existing text files into one unified document. Useful for combining outputs produced hours apart, or outputs from a model that isn't local.

Option	Description
file_a / file_b	Paths to the two manifests
--synthesis-model NAME	Model used for the merge
--output FILE	Destination file

templates

python3 prompt_expert_enhance.py templates list
python3 prompt_expert_enhance.py templates show learning_plan
python3 prompt_expert_enhance.py generate "distributed systems" --template learning_plan

See Templates and Bundles.

memory

python3 prompt_expert_enhance.py memory view
python3 prompt_expert_enhance.py memory clear

web

python3 prompt_expert_enhance.py web --port 7860
python3 prompt_expert_enhance.py web --no-browser     # headless: start server, don't open a browser

See Web UI and REST API.

---

Common options

These apply to generate, parallel, full and synthesis.

Models & sampling

Flag	Default	Description
--temperature FLOAT	0.3	Sampling temperature. 0.0 deterministic → 1.0+ chaotic. See the table below.
--timeout SECONDS	600	Per-call timeout.
--ollama-url URL	local Ollama endpoint	Point at any compatible endpoint.
--backend {ollama,openai_compatible}	ollama	Use openai_compatible for LM Studio, GPT4All server mode, or text-generation-webui's OpenAI extension. See Configuration.

Temperature guidance (from the methodology file's own system parameters):

Value	Behavior	Use for
0.0	Deterministic, reproducible	Code, extraction, classification
0.3–0.5	Balanced	Analysis, technical writing (default)
0.7–1.0	Creative, diverse	Brainstorming, fiction, exploration
>1.0	Chaotic	Stress tests, edge cases

Technique selection

Flag	Description
--techniques "1,5,8"	Specific technique IDs
--techniques "1-30"	A range
--techniques "all"	All 173
--techniques "bundle:NAME" or "bundle:3"	A pre-configured task-type bundle
--techniques "random" / "random:6"	A random subset — surprisingly good for creative unblocking
--recommend-techniques	Let the tool pick, based on your task's own wording
--list-techniques	Print all 173 grouped by category, then exit

Full catalogue: Prompt Engineering Techniques.

Output shape

Flag	Default	Description
--mode {quick,full}	full	quick = one enhanced prompt · full = 12-section manifest
--draft	off	Only sections 1–2 of a Full manifest. Fast direction check.
--quiet	off	Suppress logging — for scripting and pipes.
--output FILE	auto	Explicit destination path.

Pre-processing

Flag	Description
--no-preprocess	Skip the pre-processing step entirely; your raw text goes straight in.
--fast-preprocess	Regex-only cleanup instead of an LLM call. Much faster, no extra model round-trip.
--pre-processor-model NAME	Use a different (usually smaller/faster) model just for pre-processing.

Web enrichment

Flag	Description
--offline	Skip all web enrichment and connectivity checks. Fully air-gapped run.
--max-web-pages N	How many search results to fetch (default 1).
--summarize-web-pages	Summarize fetched pages before injecting, instead of pasting raw excerpts. Keeps the prompt tight.
--deep-research	Interactive: search, review candidate sources, pick them by hand. Needs a terminal.

Privacy & caching

Flag	Description
--anonymize	Redact personal information out of your task before it reaches any model. Regex-based — see its limits.
--no-cache	Force a fresh call even if an identical request is cached.
--no-memory	Don't feed session history into this run's context.

Templates

Flag	Description
--template ID	Start from a built-in template. If a task string is also given, it fills the template's [TOPIC] placeholder.

---

Recipes

# Fastest possible loop while wording a task
python3 prompt_expert_enhance.py generate "TASK" --mode quick --fast-preprocess --offline

# Deterministic, reproducible output for code work
python3 prompt_expert_enhance.py generate "TASK" --temperature 0.0 --techniques "bundle:Generation de code robuste"

# Maximum-quality single deliverable
python3 prompt_expert_enhance.py full "TASK" --techniques "bundle:Recherche de qualite maximale" --summarize-web-pages

# Security audit brief
python3 prompt_expert_enhance.py generate "TASK" --techniques "bundle:Audit / securite" --mode full

# Scriptable, machine-readable, no chatter
python3 prompt_expert_enhance.py generate "TASK" --mode quick --quiet --offline --output ./out.md

# Creative unblock
python3 prompt_expert_enhance.py generate "TASK" --temperature 0.9 --techniques "random:6"

Runnable versions of these live in examples/.

---

Next: Templates and Bundles · Slash Metacommands · Examples — these flags in action.