Add squad template apply command #1135

Problem

mfrieman/agentic-ai-collective
(v0.1.0-alpha) ships squad-template compositions — pre-assembled .squad/
baselines that bundle agents, skills, routing, and ceremonies scaffolding into a
single versioned package. These templates live under
compositions/squad-templates/<name>/ with a manifest.json and a scaffold/
subtree.

There is currently no upstream CLI command to apply one. Consumers have to:

squad upstream add https://github.com/mfrieman/agentic-ai-collective.git --name collective
squad upstream sync collective
cp -r ~/.squad/upstream/collective/compositions/squad-templates/general-dev/scaffold/ ./
# then manually resolve every squad:// URI in manifest.json and copy atoms

This is fragile, undiscoverable, and doesn't track what was installed.

Current state

squad upstream add / squad upstream sync can register and pull a remote
library. There is no squad template subcommand family, and no way to apply a
template composition to a new or existing project through the CLI.

Desired state

A squad template subcommand that can discover and apply template compositions
from registered upstreams or via canonical squad:// URIs.

# Discover available templates in a registered upstream
squad template list collective

# Apply a template to the current directory (new project)
squad template apply collective/general-dev

# Apply by canonical URI (pinned version)
squad template apply squad://mfrieman/agentic-ai-collective/squad-templates/general-dev@1.0.0

# Alternatively, surface this as a flag on squad init
squad init --template squad://mfrieman/agentic-ai-collective/squad-templates/general-dev@1.0.0

What apply should do:

1. Resolve the template manifest (manifest.json) from the upstream registry.
2. Copy/expand the scaffold/ subtree into the target directory.
3. Materialize all referenced atoms (agents, skills, instructions, hooks) to their
   canonical install targets (.copilot/skills/, .squad/agents/, etc.).
4. Write a .squad/assets.lock.json (or equivalent) recording source URIs,
   resolved versions, and install paths for reproducibility.
5. Skip files that already exist and match, prompt or --force for conflicts.

Acceptance criteria

• squad template list <upstream> enumerates templates from a registered
  upstream, showing name, version, and description from manifest.json.
• squad template apply <upstream>/<name> copies scaffold files and
  materializes all atoms referenced in manifest.json into the working
  directory.
• squad template apply squad://<owner>/<repo>/squad-templates/<name>@<ver>
  works without a pre-registered upstream (resolves on-demand).
• After apply, .squad/assets.lock.json (or equivalent lockfile) contains an
  entry for every materialized atom: source_uri, resolved version, and
  local install path.
• Running apply twice on the same template is idempotent: existing files that
  match the source are skipped; conflicts require --force to overwrite.
• squad template apply --dry-run prints what would be created/overwritten
  without writing any files.

References

• mfrieman/agentic-ai-collective v0.1.0-alpha — consumer library motivating this gap
• ARCHITECTURE.md — "Squad-Template Anatomy"
• ARCHITECTURE.md — "Today vs. Target" (explicit gap list)
• docs/usage/apply-a-template.md
• README.md — "Apply a squad-template"
• .squad/decisions.md line 534–535 — explicitly flags squad template list and squad template apply as upstream gaps

Notes

• Templates may target either a blank directory or an existing project. The command
  should detect whether .squad/ already exists and behave accordingly (init vs
  overlay).
• Safe extension into an existing squad (partial apply, no-clobber) is a
  scope decision for Brady; at minimum, the --force flag should make behavior
  explicit.
• This issue is intentionally scoped to the apply surface. Template authoring
  tooling (squad template scaffold, validation) is a separate concern.
• Related: issue feat: GitHub Issues intake, PRD mode, and human team members #2 (workstream add) depends on the same atom materialization
  machinery; a shared internal resolver would serve both.