| name | add-pack |
| description | Installs a ready-made agent pack into an existing team repo, checks it does not break the agents already there, and reports what it needs to work. Trigger on /add-pack, install a pack, or add the <name> pack. |
Add pack
A pack is one or more agents plus their skills, already built and audited. Installing one
should take a minute and change nothing that already works.
Before installing
Show the user three things about the pack and wait for a go:
| Show | Because |
|---|
| What it does, in one line | They should know before it lands |
| Which model each agent uses, and how often it is meant to run | It comes out of their daily allowance |
| What it needs connected | A pack that needs a CRM is useless to someone without one |
If the pack needs something they do not have, say so plainly and ask whether they still
want it installed. A pack that says "I cannot work until you connect X" on every run is
worse than no pack.
Installing
- Snapshot first.
git status must be clean. If it is not, commit or stash before
going further — a half-installed pack on top of uncommitted work is hard to unpick.
- Copy the pack's agent files into
.claude/agents/.
- Copy its skills into
.claude/skills/.
- Copy any knowledge templates into
agents/<slug>/knowledge/, fill markers intact.
- Register each new slug:
AGENT_SLUGS in scripts/lib/run-log.mjs, AGENT_SPECS in
scripts/lib/agents.mjs, IMPLEMENTED in tests/agents.test.mjs.
- Add the new specialists to
CLAUDE.md and .claude/rules/routing.md.
- Regenerate the model card:
node scripts/build-model-card.mjs.
Check — in this order
node scripts/prompt-audit.mjs
npm test
Both clean. If the audit reports findings inside the pack's own files, stop and report it —
that is a defect in the pack, not something to work around locally.
Then confirm nothing already there broke:
"Ask your research agent for anything. Then ask your content agent for anything. Both
should behave exactly as they did before."
Report back
## <Pack name> installed
**Added:** <agents>, <skills>
**Models:** <slug> on <alias>, ...
**Needs connected:** <list, or "nothing">
**Suggested schedule:** <what, how often, and what that costs against the daily cap>
Your existing agents were run after install and behave unchanged.
Uninstalling
Removal is the reverse, and it has to leave the repo exactly as it was:
- Delete the pack's files from
.claude/agents/ and .claude/skills/
- Remove its slugs from the three registration points
- Regenerate the model card
npm test
Run logs the pack's agents wrote stay. They are history, and history is not undone by an
uninstall.
