Llm

Build structured prompts, estimate tokens, compare prompt variants, and manage reusable prompt templates.

Security Scan

The skill is internally consistent with its stated prompt-engineering purpose: it is an instruction-only bash toolkit that builds/compares/tokenizes prompts and manages templates in the user's home directory, and it requests no external credentials or unusual system access.

Purpose & Capability

Name/description (prompt building, token estimation, comparing variants, templates) match the delivered files: SKILL.md documents those commands and scripts/script.sh implements prompt, compare, tokenize, template, chain, and evaluate behaviors. No unrelated services, credentials, or exotic binaries are required.

Instruction Scope

Runtime instructions tell the agent to run the bundled bash script with specific subcommands and options. The script operates on provided files, stdin, and a template directory (~/.llm-skill/templates). There are no instructions to read unrelated system files, environment secrets, or to send data to external endpoints.

Install Mechanism

No install spec is provided (instruction-only), lowering risk. The included code is a bash script that runs locally; it does not download or extract remote artifacts. It relies on common POSIX tools (cat, wc, diff, mkdir, etc.), which is proportional to its function.

Credentials

The skill declares no required environment variables or credentials. The script writes templates under the user's home directory but does not attempt to read or exfiltrate secrets or access unrelated config paths.

Persistence & Privilege

The skill is not always-enabled and does not request elevated privileges. Its only persistent side-effect is creating and managing files in ~/.llm-skill/templates and logging minor messages to stderr — behavior consistent with a template manager.

Assessment

This skill runs a local bash script (requires bash 4+) and will create/manage files under ~/.llm-skill/templates. It does not request credentials or make network calls in the provided code. Before installing or running, review the script if you want to be extra cautious, and be aware it can overwrite/delete template files you name (template --save/--delete). If you run it in a multi-user or sensitive environment, consider running it under a non-privileged user or inside an isolated environment (container) to limit accidental file changes.

LLM prompt engineering toolkit. Build structured prompts, compare variations, estimate tokens, manage templates, chain multi-step prompts, and evaluate prompt quality.

Commands

prompt

Build a structured prompt from role, context, and task components.

bash scripts/script.sh prompt --role "senior developer" --context "Python Flask app" --task "write unit tests"

compare

Compare two or more prompt variations side by side with diff output.

bash scripts/script.sh compare --prompts prompt_a.txt prompt_b.txt

tokenize

Estimate token count for a given text using cl100k_base-compatible counting.

bash scripts/script.sh tokenize --input "Your prompt text here"
bash scripts/script.sh tokenize --file prompt.txt

template

Manage reusable prompt templates: save, list, load, and delete.

bash scripts/script.sh template --save my_template --file prompt.txt
bash scripts/script.sh template --list
bash scripts/script.sh template --load my_template
bash scripts/script.sh template --delete my_template

chain

Define and run multi-step prompt chains where each step feeds into the next.

bash scripts/script.sh chain --steps step1.txt step2.txt step3.txt
bash scripts/script.sh chain --from chain_config.json

evaluate

Score prompt quality based on clarity, specificity, structure, and completeness.

bash scripts/script.sh evaluate --input "Your prompt text"
bash scripts/script.sh evaluate --file prompt.txt

Output

All commands output to stdout in plain text or JSON (with --json flag). Token counts return integers. Evaluate returns a 0-100 score with breakdown. Templates are stored in ~/.llm-skill/templates/.

Requirements

• bash 4+

Feedback

Report issues or suggestions: https://bytesagain.com/feedback/

---

Powered by BytesAgain | bytesagain.com