Use Cases

Use Cases — what people actually build with Wild_Root_Prompt

Wild_Root_Prompt is a local prompt engineering tool, but that phrase hides what it's for. Concretely: it turns an underspecified request into a brief good enough that a model — local or hosted — stops guessing.

Each use case below names the job, the command, and the honest limit.

---

1. Writing prompts for other LLMs

The job: you're going to paste this into a hosted model. You want the prompt to be worth the tokens.

python3 prompt_expert_enhance.py generate "YOUR TASK" --mode quick

Quick mode outputs one enhanced prompt, ready to paste anywhere. This is the highest-frequency use of the tool, and the reason it works offline: the thinking is local even when the execution isn't.

See it happen: Examples.

---

2. Briefing an AI coding agent

The job: an agent will act on your instructions across many steps, and vague instructions compound into wasted work.

python3 prompt_expert_enhance.py generate "YOUR TASK" --mode full \
  --techniques "bundle:Generation de code robuste"

The 12-section manifest exists for exactly this: ambiguity zones surfaced up front, guardrails stated, error handling defined, and a reproducibility checklist at the end. It's written to be executed, not admired.

---

3. Learning a subject you can't yet name

The job: you know what hurts, not what to ask. Beginners lose most of their time here.

python3 prompt_expert_enhance.py generate "how do i make my python script faster it takes forever" \
  --mode quick --techniques "bundle:Apprentissage / explication"

The pre-processor supplies the vocabulary you were missing — see Example 2, where "it takes forever" turned into profiling, cProfile, and benchmarking. Add /niveau:debutant to calibrate depth.

---

4. Technical specifications and design docs

The job: a spec that survives review — goals and non-goals, alternatives considered, edge cases, rollout.

python3 prompt_expert_enhance.py generate "OAuth2 device flow" \
  --template technical_spec --mode full

The technical_spec template already contains that skeleton, so the model starts with structure instead of inventing one.

---

5. Security and code review checklists

The job: find what you'd miss by reading sympathetically.

python3 prompt_expert_enhance.py generate "YOUR SYSTEM" \
  --techniques "bundle:Audit / securite" --mode full

That bundle is red team/blue team, stress testing, assumption mapping, bias hunting and invariant detection. Stack /critique /risques /hypotheses for an unsympathetic pass.

---

6. Decisions with real consequences

The job: choose, and be able to defend the choice later.

python3 prompt_expert_enhance.py full "Postgres or DynamoDB for this workload" \
  --techniques "bundle:Decision critique"

Tree-of-Thought, pre-mortem, steelmanning, assumption mapping and dual-process forcing — the combination that makes a model argue against the option it likes. Run it as full so two models argue and a third synthesizes.

---

7. Comparing two models honestly

The job: you want to know which local model is better at your work, not on a leaderboard.

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

Split-screen, same prompt, same techniques, live tokens. Then let synthesis merge the best of both — see Architecture.

---

8. Course and curriculum design

python3 prompt_expert_enhance.py generate "distributed systems" --template create_course
python3 prompt_expert_enhance.py generate "rust" --template learning_plan

Syllabus, weekly modules, learning objectives, assessments, capstone. Add /niveau:debutant or /niveau:expert to pitch it correctly.

---

9. Long-document synthesis

python3 prompt_expert_enhance.py generate "YOUR SOURCE MATERIAL" \
  --techniques "bundle:Synthese de document long"

Skeleton-of-Thought, recursive summarization, outline-then-expand, priority stacking and MECE — built to resist the lost in the middle failure mode where a model quietly drops the middle of a long input.

---

10. Structured, machine-readable output

python3 prompt_expert_enhance.py generate "/json /silence extract these fields" \
  --techniques "bundle:Output parseable / API" --temperature 0.0

Format forcing, strong delimiters, constraint stacking, conditional prompting — plus temperature 0.0 for determinism. Drive it from your own program via the REST API.

---

11. Air-gapped and regulated environments

The job: you cannot send this text to a third party. Not "prefer not to" — cannot.

python3 prompt_expert_enhance.py generate "YOUR TASK" --offline

--offline skips web enrichment and connectivity checks entirely. Generation was always local. There are no API keys in the project because there is nothing to authenticate to.

---

12. Breaking a creative block

python3 prompt_expert_enhance.py generate "YOUR TOPIC" \
  --techniques "random:6" --temperature 0.9

A random technique subset is a genuinely effective unblocking device — it forces angles you wouldn't have chosen. Or use the Deblocage creatif bundle: inversion, counterfactuals, alien-anthropologist framing.

---

Who it fits

You are	Start here	Why
Not a terminal person	Web UI	One click, browser, done
A developer scripting it	CLI Reference	Flags, bundles, --quiet --output
Building on top of it	REST API	SSE streaming, six endpoints
Studying prompt engineering	The 173 techniques	A categorized catalogue with anti-patterns
On constrained hardware	Installation	A 3B model in ~4 GB works; Android via Termux

---

Where it is the wrong tool

Being straight about this saves you time:

• You want a chat interface. This produces prompts and manifests. Use a chat client for chatting.
• You want verified facts. It structures reasoning; it does not fact-check. See Limitations and Roadmap.
• You want one-word answers. The techniques push toward depth. For terse output use /concis or /minimal — but a simpler tool may suit you better.
• You have no local model and don't want one. Ollama or an OpenAI-compatible local server is required by design.

---

Next: Quick Start · Examples · Comparison