Grok Bot Local First (v0.1.0)

    1

    Local-first skill: keep Grok Bot turns for judgment only; deterministic work runs in Python 3.12 CLI workers that emit last-line JSON.

    Free

    0 installsSecurity scanned

    Works with the AI tools you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIGitHub CopilotGitHub CopilotGemini CLIGemini CLI+17 more

    See it in action

    You say

    Ingest the worker in the drop folder if present, then run only its default or lowest progressive action (e.g. --action d1 or --action r1). Report the last-line JSON. Do not read worker source. Do not climb higher levels.

    Your agent does

    Ingested. Running: python disclose-opencode-muse-spark.py --action d1 Last line: {"ok":true,"alert":false,"summary":"identity + tiny JSON note","data":{"provider":"muse-spark","ui":"opencode","note":"emit tiny last-line JSON"},"action":"d1"} use_bot is false / alert is false. Staying quiet; no further bot turn needed.

    About this skill

    Early research (week-horizon)

    This skill captures a working local-first pattern developed against Grok Bot 0.27.x. It is intentionally Grok-optimized and still evolving. Fork and adapt freely; the northern star is stable, the examples are not permanent.

    The problem

    Every bot wake is expensive. Putting file loops, parsing, or research packing into the model wastes allotment and floods context. Script source and raw data should almost never sit in the prompt.

    What it does

    • Routes the bot to one --action on a local Python 3.12 stdlib worker.
    • Reads only the last stdout line (tiny JSON: ok, alert, summary, data, action).
    • Uses progressive flags (d1dN or r1rN) so the bot climbs one level per turn.
    • Keeps description and memory-profile tiny; procedures live in workers or in references/ (loaded on demand).
    • Lets external models (web Grok 4.5 Fast, Claude, GPT, OpenCode families, etc.) write workers from a pure CLI contract—no harness internals required.

    How the split works

    Bot = router + judgment. Worker = deterministic work. External research partner (e.g. web Grok 4.5 Fast) writes the worker or a progressive research pack; the bot only runs flags and inspects JSON. The same contract has already been exercised with multiple model families.

    Why this beats ad-hoc prompting

    Without a rigid contract, models re-implement work, invent schemas, or dump large intermediate text into context. This skill forces the opposite: last-line JSON, one action, stay quiet when quiet: true or use_bot: false.

    Use cases

    • Live request that can be answered by an existing worker → run flag, no bot loop.
    • Recorded / scheduled tasks that only page the bot on alert: true or ok: false.
    • Research packs that climb r1rN without loading the full result “just in case.”
    • Cross-model worker authorship: give other models only the CLI contract; stamp public provider + ui.

    Known limitations (honest)

    Early research state. Requires a local runner that can register workers from a drop folder and execute python worker.py --manifest|--list-actions|--action NAME. Tuned for Grok Bot 0.27.x; other agents can adopt the same SKILL.md layout and CLI contract. Workers are Python 3.12 stdlib only. Not a general task scheduler or multi-language runtime.

    How to install

    Drop the file into your AI Agent. Works with Claude, Cursor, ChatGPT, and 20+ more.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill passes our trust checks below.

    Security scanned

    Passed our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scanned

    Verified clean today

    • Free to download with an account

    Listedtoday

    Frequently Asked Questions

    Free