How to use the Open Source Repo Manager agent skill

What is an agent skill and what does this one do?

An agent skill is a folder of instructions and scripts that an AI agent loads when a request matches the skill's purpose. This one is scoped to the list of repositories you configure in a Config.md file, under the keys OSS_OWNER and OSS_REPOS. Repositories outside that list are out of scope.

Through the gh command-line tool, the agent fetches live data and reports on it. It ranks items by urgency and impact, rates each repo's community mood as positive, neutral or negative, and suggests specific replies. It is built to stop and wait for your go-ahead before any public change, such as merging, publishing or pulling a model.

• Triage open issues and pull requests, with suggested replies
• Read GitHub Discussions and surface unanswered Q&A
• Post or update one marker comment on an issue or PR instead of stacking new ones
• Diff newly published Ollama library models against a local model host
• Run a readiness audit on a private repo and build a clean fresh-history export

Install the OpenSourceManagement skill

Download the skill ZIP from this page and unzip it into one of two places. For all your projects, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. The folder should be named OpenSourceManagement.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside the project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up the requirements before first use

The skill needs the Bun runtime, the GitHub CLI (gh) installed and logged in with access to the repositories you want to triage, and a Config.md file that lists OSS_OWNER and OSS_REPOS. The Ollama check is optional and needs an Ollama host you can reach over ssh. The skill's Setup.md file covers first-time configuration and includes a proof command you can run to confirm the setup works.

Config.md lives under LIFEOS/USER/CUSTOMIZATIONS/SKILLS/OpenSourceManagement/. Without the gh login and the config file, the skill cannot run.

• Bun runtime
• GitHub CLI (gh), logged in, with access to your repos
• Config.md with OSS_OWNER and OSS_REPOS (optional OSS_PRIORITY)
• Optional: an Ollama host reachable by ssh, set in the Local Model Host section of Config.md

Trigger the skill with natural requests

You do not need to call the skill by name. The skill is chosen when your request matches its trigger phrases. Ask in plain language and say which repo or area you mean. The skill routes each request to one of seven intents: issues, pull requests, discussions, a state snapshot, the marker comment, local models, or going public.

If you name a single repo, the fetch is narrowed to that repo. If you name one that is not in your list, the skill will not triage it.

• "Check open issues across my open source repos"
• "Review PRs on <repo> and tell me which to merge"
• "What are people asking in community discussions?"
• "What needs attention in our open source projects?"
• "Post a triage comment on issue 42 in <repo>"
• "What new open-weight models should I pull into Ollama? Do we already have this one?"
• "Is <repo> public-ready?"

Run a triage and read the report

A triage report is built from a fresh fetch in the current session, not from memory. Each repo gets its own section. Items are listed by priority with a category, a sentiment for the project, and a suggested response. If a fetch failed, for example because gh lacked the right scope, the report names that repo as a gap instead of showing it as quiet.

Lists are capped. The issues and PR fetch shows the first 50, and discussions show the 30 most recently updated. When a list is truncated, the report says so.

• Priority per item, with suggested replies you can edit before posting
• Hole notice when a repo could not be fetched
• Truncation notice when a list hit its limit

Use the marker comment for recurring bot notes

When the skill posts a status note, it uses hidden HTML markers so the same comment is edited in place on each run rather than adding a new one. The skill can also find an existing marker comment on an issue or PR. This keeps a thread clean when you check it repeatedly.

• Create or update one comment per marker ID
• Dry-run mode previews the change before it is posted

Check new Ollama models against your local box

Ask the skill which new open-weight models are in the Ollama library and whether they fit your hardware and are already on your local model box. The host comes from the Local Model Host section of Config.md. The skill reports what is new and leaves every pull to you.

The skill does not run or prompt the models. Running local models is outside its scope.

• Asks for newest library models, compared with what is installed
• Can be limited to a number of results or run against the local machine

Take a private repo public safely

For a private repo, ask whether it is public-ready. The skill runs an audit of its readiness checks and can produce a clean export with fresh history into an empty directory. The export does not push anything. The skill also covers the README and a social preview image, and it waits for your explicit go-ahead naming that repo before any public change.

Repos that are generated or synced from another source tree are handled differently. The skill points changes at the source of truth in Config.md, so a fix made only on the public copy is not lost on the next sync.

• Run an audit that reads READY or lists problems
• Export a fresh-history copy to an empty folder (never pushes)
• Wait for your go-ahead before making anything public

Safety notes: scripts and network access

This skill runs scripts and uses the network. The fetch and comment tools call GitHub through gh, and the model check can connect to an ssh host. The skill is designed to avoid public changes without your approval: it does not merge, publish or pull models on its own. Review any suggested reply or comment before it goes out, and use the dry-run option for comment upserts when you are unsure.

• Scripts run locally with Bun and call gh
• Network access goes to GitHub and, if configured, your Ollama host over ssh
• Public actions wait for your go-ahead