snqb/scripting

Scripting Defaults

Decision

| Script needs… | Use | Skill to load | |---|---|---| | Shell commands, pipes | Deno + dax | deno | | HTTP/data/batch | Deno + dax | deno + longrun | | Python-only lib (Telethon, bip_utils, libgen-api, radon…) | Python + uv run | python |

Default is Deno. Only pick Python when a required library doesn't exist in JS/TS.

Deno (default)

#!/usr/bin/env -S deno run --allow-all
import $ from "jsr:@david/[email protected]";
import { pooledMap, retry } from "jsr:@std/async";
import { sortBy, partition, chunk, sumOf } from "jsr:@std/collections";

Single file, inline jsr: imports, no config files. Load deno skill for full reference.

Python (fallback)

#!/usr/bin/env -S uv run
# /// script
# requires-python = ">=3.12"
# dependencies = ["httpx", "plumbum"]
# ///
from plumbum import local, RETCODE
from plumbum.cmd import git, curl, jq

PEP 723 inline deps, zero-setup. plumbum for shell commands (only lib where piping works). Load python skill for full reference.

Never generate venvs, requirements.txt, or deno.json for throwaway scripts.

pi-llm — LLM Instead of Heuristics

Before writing fuzzy matching, entity resolution, or classification logic — use ask(). It's cheaper than iterating on string algorithms.

| Task | ❌ Don't | ✅ Do | |------|---------|-------| | "Same hotel?" | Jaccard, identity keywords, compound tokenizer (10 iterations) | ask("Same hotel? A='X' B='Y'") — done, first try | | "What category?" | regex chains, keyword lists | ask("Classify: ...") | | "Extract field" | regex with 12 edge cases | ask("Extract room type from: ...") |

import { ask, run } from "~/.pi/agent/lib/pi-llm.ts";

const answer = await ask("Is this critical?", { model: "claude-haiku-4-5" });       // instant judgment
const result = await run("Research X", { tools: "full", maxTurns: 8 });              // agent with tools

Rule: if you're on iteration 2+ of fixing edge cases in string matching, you already wasted time — should have used ask() from the start. Tokens are unlimited (Max subscription). Optimize for results, not cost. See longrun skill for batch patterns.