Labeled-Issues-Based-Execution (LIBE) for GitHub — Turn GitHub Issues into deterministic workflows for AI agents.
Status: Work in progress. Not ready for production use.
libe-coding maps GitHub Issue labels to scripts and agentic workflows. Instead of hoping an AI gets the task right, you give it a verified call to action so every run follows the same recipe. When a label matches, the right context is applied to the task.
“I thought GitHub Copilot could do this, but the power of Cursor and Claude’s agentic programming is way better.”
I know there are already tools like this out there, but I wanted to create a tool that uses a proven issues process for teams and can be used locally on my machine and in GitHub workflows.
-
Label-triggered execution — Match issue labels to action folders; each action folder’s
execute.shruns when the label is applied. It can also use the context, and if an agentic workflow is used it will use your AGENTS.md / CLAUDE.md, tools, and skills as configured for your agent. -
Deterministic workflows — Reusable prompts and scripts instead of ad-hoc instructions.
-
Flexible scheduling — Run once or on a cron interval (e.g. every 15 minutes).
-
Run in GitHub workflows — Use the libe-coding package in GitHub Actions workflows.
- Node.js (v22+)
- GitHub token — Create a Personal Access Token with
reposcope. Set it asGH_TOKENin your environment or config.
git clone <this-repo>
cd libe-coding
npm install
npm run buildYou can run libe-coding without installing it globally:
npx libe-coding start --once --repo owner/my-repoSet GH_TOKEN in your environment (or use gh auth login) before running.
-
Set your token
export GH_TOKEN=ghp_xxxx -
Run once (e.g. for repo
owner/my-repo)npm start -- --once --repo owner/my-repo
-
Run on a schedule (default: every 15 minutes)
npm start -- --repo owner/my-repo
Config can come from:
- CLI options (see below)
- Config file —
.libecodingrc,.libecodingrc.json, orlibe-sample.json(via cosmiconfig) - Environment — e.g.
GH_TOKEN
Hint: Copy libe-sample.json to .libecodingrc and set your repo and token (or rely on GH_TOKEN).
| Option | Alias | Default | Description |
|---|---|---|---|
--config-folder |
-c |
libe |
Base config folder; actions are under <config-folder>/labels/ |
--history-folder |
— | libe/history |
Folder for execution history |
--repo |
-r |
— | GitHub repo (e.g. owner/name) |
--interval |
-i |
*/15 * * * * |
Cron expression (default: every 15 min) |
--once |
— | false |
Run once and exit (ignores interval) |
--verbose |
-v |
false |
Print more details (action paths, per-issue labels, match/no-match results) |
--help |
-h |
— | Show help |
Labels are matched to directory names under <config-folder>/labels/. Matching is case-insensitive. The app appends labels to your config folder, so with default libe it uses libe/labels/.
Example: label libe-improveissue or libe-ImproveIssue runs the action in libe/labels/libe-ImproveIssue/.
Tip: Use a prefix like libe- to avoid clashing with other labels.
Put one folder per label under <config-folder>/labels/ (default: libe/labels/). Each action folder should contain:
execute.sh— Script that runs when the label is applied. Receives:$1= path toCONTEXT.MD,$2= matched label name.CONTEXT.MD— Instructions and context for the action (and for the script).
Environment variables passed into execute.sh:
ISSUE_TITLE,ISSUE_BODY,ISSUE_NUMBERREPO,REPO_OWNER,REPO_NAMEMATCHED_LABEL— Label that triggered this action (same as$2)
Example layout (default config folder libe → actions under libe/labels/):
my-repo/
├── libe/
│ ├── labels/
│ │ ├── libe-ImproveIssue/
│ │ │ ├── execute.sh
│ │ │ └── CONTEXT.MD
│ │ └── libe-CreateTest/
│ │ ├── execute.sh
│ │ └── CONTEXT.MD
│ └── history/
│ ├── 2026-02-01T12-30-00-123Z_42.md
│ └── 2026-02-01T13-15-00-456Z_43.md
├── libe-examples/
│ └── labels/
│ └── libe-test/
│ ├── execute.sh
│ └── CONTEXT.MD
├── package.json
└── .libecodingrc
Every action execution is automatically logged to the history folder (default libe/history). Each history file is named {timestamp}_{issue_number}.md and contains:
- Timestamp, issue number, title, and repo
- Matched label and execution status (success/failure)
- Issue body and link to the GitHub issue
- Error details (if the execution failed)
History folders are listed in .gitignore so execution logs stay local.
This repo includes example actions under libe-examples/labels/. To use them, set the config folder to the base so the app uses libe-examples/labels/:
npm start -- --once --repo owner/my-repo -c libe-examplesFor hints and your own actions, use the libe folder (not libe-examples). Hint: To get started, copy libe-examples to libe (e.g. cp -r libe-examples libe); then put your action folders under libe/labels/ and run with the default config. You can add libe (or a custom folder) to .gitignore if you want to keep prompts and scripts private.
Run once for a specific repo
npm start -- --once --repo my-org/my-repoVerbose output (action paths, per-issue labels, match results)
npm start -- --once --repo my-org/my-repo -vCustom config folder
npm start -- --once --repo my-org/my-repo -c libe-examplesCustom cron interval (e.g. every hour)
npm start -- --interval "0 * * * *" --repo my-org/my-repoConfig file
Copy libe-sample.json to .libecodingrc (or .libecodingrc.json), then set your repo and token:
{
"repo": "owner/my-repo",
"configFolder": "libe"
}Actions go under libe/labels/ (e.g. libe/labels/libe-test/).
Then run:
npm startAdd a comment to an issue (comment-issue)
The comment can be markdown or plain text. Provide it via --body, --file, a positional argument, or stdin. Repo and issue: -r owner/repo, --issue <number> (or env REPO, ISSUE_NUMBER).
Use npm run libe -- when developing locally; use npx libe-coding once the package is published.
The idea came from wanting to keep coding moving while away from the keyboard — for example on a long walk in the snow. AI agents are powerful but inconsistent: the same task can work once and fail the next time. libe-coding applies DRY to agent workflows: you define the recipe once (scripts + context), and the agent runs it every time a labeled issue is picked up. You manage work via GitHub Issues; the agent gets a clear, repeatable playbook.
MIT
