skill-loader

by João Savi · ★ 0 · updated 4 days ago

GuardsPromptAutomationwarnings

0

Loads the right skill at the right moment, so CLAUDE.md needs no load-this-skill rules: typescript-best-practices for .ts, writing-for-agents for SKILL.md, AGENTS.md and CLAUDE.md, domain-modeling for CONTEXT.md and ADRs, technical-writing for READMEs, commits and PRs, and a diagnosing-bugs reminder when you paste an error.

install skill-loader

Install

/plugin marketplace add joaosavi/cockpit
/plugin install skill-loader@cockpit

Then run /reload-plugins or start a new session. Requires Claude Code 2.1.287+.

View source on GitHub

What it hooks into

From an automated scan with claude plugin validate, not a security review. Mods run with your permissions; read the code before installing.

Events
session.starttool.callprompt.submitsession.compactsession.end
API calls
$.clock.now$.env.get$.fs.read$.fs.write$.session.cwd$.session.id
Can see
every tool callevery promptcompaction
Reach
level 2 of 3 writes filesreads filesreads env vars

Origin

How it got here
Automated scan of public GitHub repos, via the CC0 dataset awesome-claude-code-mods (scan of ).
Repository
github.com/joaosavi/cockpit
Path in repo
skill-loader
Source commit
7edcf86 the exact code the scan read
Author
João Savi @joaosavi
License
MIT
First published
Last pushed
First listed here

Name and description come from the mod's own manifest. We link to the code; we don't host it, and it stays under its author's license.

More from joaosavi/cockpit

collision-guard

joaosavi/cockpit

Warns when two Claude Code sessions on this machine edit the same file, and holds a git command once when it would stage, commit, stash or discard another session's uncommitted edits.

★ 0ToastsGuards

global-memory

joaosavi/cockpit

Adds a machine-wide memory folder beside the per-project auto memory, so facts about the machine reach every session from any folder: rewrites the memory section for both folders and loads the global MEMORY.md.

★ 0PromptAutomation

hud

joaosavi/cockpit

A heads-up display under or above the prompt: model, branch, context and plan limits, and memory headroom. /hud opens the full view

★ 0PanesBandsGuards

loop-breaker

joaosavi/cockpit

Notices when the agent goes in circles (repeated fix attempts on one file, the same error coming back, the same call repeated) and tells it to stop and change approach; refuses nothing.

★ 0GuardsAutomation

memory-guard

joaosavi/cockpit

Keeps typecheck, build, test and Playwright runs under the memory limit, the tightest cgroup limit on Linux or the machine's memory on Linux, macOS and Windows: runs them with one process, waits for memory or refuses, and reruns a command killed with exit 137.

★ 0PanesStatus lineGuards

panel

joaosavi/cockpit

A local web page with a card for every Claude Code session on this machine: whether it waits for you, what it needs, what it is doing, with a summary from claude -p every few minutes and a refresh button.

★ 0CommandsAutomation

pipefail

joaosavi/cockpit

Adds set -o pipefail to a command that pipes a test, typecheck or lint run into tail, head or grep, so a failing run cannot exit 0 and read as a pass.

★ 0GuardsAutomation

secret-guard

joaosavi/cockpit

Hides every secret value in tool output as before the agent reads it, MCP and subagent output included, and writes the real value back when the agent puts the placeholder into a .env or secrets file.

★ 0GuardsAutomation

spinner-status

joaosavi/cockpit

Replaces the tip under the spinner with one short line from Haiku: where this session stands, or a tip about your mods and their commands that fits the moment

★ 0Reskins

todo

joaosavi/cockpit

Keeps the native task list in front of the agent: the open tasks at session start and right after each compaction, a reminder that a new request becomes a task before any other work, completed tasks logged to an archive, and the app's task panel kept in line with the task files.

★ 0GuardsPromptAutomation

usage-limit

joaosavi/cockpit

Keeps the account usage limit from running out mid-task, with separate limits for the 5-hour and the 7-day windows: one subagent at a time (90% and 97%), a commit of the work in progress (97% and 98%), a pause that keeps the rest for you (98% and 99%), each one switchable, and an automatic resume when the window resets.

★ 0BandsStatus lineGuards