agentic-workshop

Skills for running coding agents as a team.

foreman-start

One agent acts as the foreman over a markdown TODO checklist. Every open item gets its own paseo workspace: a git worktree off the default branch, with its own agent. At most two workspaces run at a time.

Each workspace does the whole job itself:

• implement the item;
• prove it with a test that first fails on the old code;
• run the project's fast check once, when the diff is ready for review;
• get a review subagent to review the diff;
• run the pre-merge check once, on the tree merged with the default branch;
• merge through a PR.

The foreman:

• plans the order by file conflict, so parallel workspaces never edit the same files;
• reads the project's verification stages and names the right command in every prompt, so each stage runs once, when it proves something new;
• writes self-contained prompts;
• gives each workspace the paseo agent profile whose notes fit the item;
• watches with heartbeats, catching missed signals, pending permission requests, usage-limit stalls and daemon restarts;
• checks each merge itself;
• archives the workspace, ticks the item off with its PR number, and reports to you;
• runs the deploy-level check (end-to-end tests, recordings) once at the end of the cycle.

It keeps its state in a comment block at the top of the TODO file. So /foreman-start also resumes a cycle after a restart or a lost context.

Requirements

• The paseo MCP server, with workspaces, agents and heartbeats.
• git with a remote, and the GitHub CLI gh, authenticated.
• Optional: paseo agent profiles, each with notes on what it is for. Without them, the workers copy the foreman's own settings.
• Optional: a merge-to-main skill in the workers' client. Without it, the workers merge with plain gh.

Install

The two copies hold the same procedure. The Claude Code copy uses Claude's tool names. The other follows the general Agent Skills layout (.agents/skills/), which Kimi Code and other clients read.

# Claude Code, for all projects
cp -r .claude/skills/foreman-start ~/.claude/skills/

# Kimi Code or another Agent Skills client, for all projects
mkdir -p ~/.agents/skills && cp -r .agents/skills/foreman-start ~/.agents/skills/

For a single project, copy the folder into that project's .claude/skills/ or .agents/skills/ instead. Start a new session afterwards so the skill is picked up.

Use

Write a checklist, for example .scratch/TODO.md:

# Controls
- [ ] Make J fire the cannon
- [ ] Show an icon per colour in the key legend

# Weather
- [ ] Add a storm weather type with lightning

Then run /foreman-start, or /foreman-start path/to/TODO.md. Without a path, it takes the newest TODO*.md with open items, looking in .scratch/ first and then the repo root.

While it runs you can:

• add items to the list;
• answer the decisions it brings up;
• ask it to deploy. It never deploys on its own.

jira-cli

Gets an agent working with Jira through jira-cli, the jira command. The skill:

• checks that jira is installed, and installs it on request: Homebrew, Scoop, or on Linux the newest release binary with its checksum verified;
• checks the login with jira me, and walks you through the API token and jira init in your own terminal, so the token never passes through the chat;
• makes the agent read the jira-cli authors' own agent rules (llm.md) before its first command;
• picks the right Jira project when you have more than one;
• downloads ticket attachments such as screenshots, which neither jira-cli nor Atlassian's acli can.

Needs curl (Linux install, attachments), python3 (attachments), and a Jira API token. Install it like foreman-start above, with jira-cli as the folder. Then ask about a ticket, or run /jira-cli.

llm.md is an unmodified copy of upstream's, taken at commit e74646e (22 Sep 2026). It is kept in the skill so nothing depends on GitHub at run time. To refresh both copies, then update the commit here:

for d in .claude .agents; do gh api repos/ankitpokhrel/jira-cli/contents/llm.md \
  -H 'Accept: application/vnd.github.raw' > $d/skills/jira-cli/llm.md; done

License

MIT, see LICENSE. The llm.md in the jira-cli skill is © 2022 Ankit Pokhrel, also MIT, copied from jira-cli.