Update Swain

Update the local installation of swain skills to the latest version, then reconcile governance configuration.

Step 1 — Detect current installation

Check whether .claude/skills/ contains any swain-* directories:

ls -d .claude/skills/swain-* 2>/dev/null

If no swain skill directories are found, inform the user this appears to be a fresh install rather than an update, then continue anyway — the steps below work for both cases.

Step 2 — Backup local modifications

Before overwriting skill directories, check for user modifications that should be preserved.

Detect modified files

Compare local skill files against the installed version's git origin:

# Create a temporary reference copy
tmp=$(mktemp -d)
git clone --depth 1 https://github.com/cristoslc/swain.git "$tmp/swain" 2>/dev/null

# Find locally modified files
modified_files=()
for skill_dir in .claude/skills/swain-*/; do
  skill_name=$(basename "$skill_dir")
  ref_dir="$tmp/swain/skills/$skill_name"
  [ -d "$ref_dir" ] || continue

  while IFS= read -r file; do
    rel="${file#$skill_dir}"
    ref_file="$ref_dir/$rel"
    if [ -f "$ref_file" ]; then
      if ! diff -q "$file" "$ref_file" >/dev/null 2>&1; then
        modified_files+=("$file")
      fi
    else
      # File exists locally but not in upstream — user-added file
      modified_files+=("$file")
    fi
  done < <(find "$skill_dir" -type f)
done

Backup modified files

If modified files are found:

1. Create a backup directory: .agents/update-backup/<ISO-date>/
2. Copy each modified file preserving directory structure
3. Inform the user: "Found N locally modified files — backed up to .agents/update-backup/<date>/"
4. List the modified files

If no modified files are found, skip and continue.

The reference clone from this step can be reused as the fallback source in Step 3.

Step 3 — Detect installed agent platforms

Before installing, detect which agent platforms are present on the system. This avoids creating dotfolder stubs for every supported platform (see GitHub issue #21).

Read the agent platform data from references/agent-platforms.json. Each entry in the agents array has a name (skills CLI identifier), an optional command (CLI binary), and a detection path (HOME config directory). A platform is detected if either check succeeds. Entries in always_include are added unconditionally.

SKILL_DIR="$(find . .claude skills -path '*/swain-update/references' -print -quit 2>/dev/null | sed 's|/references$||')"
detected_agents=()

# Always-include platforms (we're running inside claude-code)
for name in $(jq -r '.always_include[]' "$SKILL_DIR/references/agent-platforms.json"); do
  detected_agents+=("$name")
done

# Detect remaining platforms via command -v or HOME dotfolder
while IFS= read -r entry; do
  name=$(echo "$entry" | jq -r '.name')
  cmd=$(echo "$entry" | jq -r '.command // empty')
  det=$(echo "$entry" | jq -r '.detection // empty')

  # Skip always-include (already added)
  for ai in "${detected_agents[@]}"; do [[ "$ai" == "$name" ]] && continue 2; done

  found=false
  if [[ -n "$cmd" ]] && command -v "$cmd" &>/dev/null; then
    found=true
  fi
  if [[ -n "$det" ]] && ! $found; then
    det_expanded=$(echo "$det" | sed "s|~|$HOME|g")
    det_expanded=$(eval echo "$det_expanded" 2>/dev/null)
    [[ -d "$det_expanded" ]] && found=true
  fi

  $found && detected_agents+=("$name")
done < <(jq -c '.agents[]' "$SKILL_DIR/references/agent-platforms.json")

Build the -a flags from detected agents:

agent_flags=""
for agent in "${detected_agents[@]}"; do
  agent_flags="$agent_flags -a $agent"
done

Tell the user which platforms were detected:

Detected N agent platform(s): claude-code, codex, gemini-cli, ...

Step 4 — Update via npx

Run the skills package manager with only the detected agents:

npx skills add cristoslc/swain $agent_flags -s '*' -y

This installs all skills (-s '*') for only the detected platforms, skipping confirmation (-y). No dotfolder stubs are created for platforms that aren't installed.

If npx fails (command not found, network error, or non-zero exit), fall back to a direct git clone:

tmp=$(mktemp -d)
git clone --depth 1 https://github.com/cristoslc/swain.git "$tmp/swain"
# Detect skill install location
INSTALL_DIR=$(find . -maxdepth 2 -name "swain-doctor" -type d -print -quit 2>/dev/null | sed 's|/swain-doctor$||')
INSTALL_DIR="${INSTALL_DIR:-.claude/skills}"
cp -r "$tmp/swain/skills/"* "$INSTALL_DIR/"
rm -rf "$tmp"

Step 5 — Reconcile governance

Invoke the swain-doctor skill. This validates governance rules, cleans up legacy skill directories (including any renamed in this release), validates .tickets/, and untracks any runtime files that leaked into git. The skill is idempotent, so running it after every update is always safe.

Step 6 — Restore guidance

If files were backed up in Step 2:

1. List the backed-up files with their paths
2. For each, explain the situation:
   • User-added config files (e.g., config/yazi/yazi.toml): Suggest moving to .agents/config/<skill-name>/ or swain.settings.json where they'll survive future updates
   • Patched scripts: Show the diff between the backup and the new version. If the upstream version includes the fix, confirm the patch is no longer needed. If not, offer to re-apply the patch.
3. Remind the user: "To avoid this in future, store customizations in .agents/config/ or swain.settings.json — these survive updates."

Step 7 — Report

Display the current release version (from the latest git tag):

release_tag=$(git tag --sort=-v:refname | head -1)
echo "swain ${release_tag:-(unreleased)}"

Then list the installed swain skill directories and extract each skill's version from its SKILL.md frontmatter:

for skill in .claude/skills/swain-*/SKILL.md; do
  name=$(grep '^name:' "$skill" | head -1 | sed 's/name: *//')
  version=$(grep 'version:' "$skill" | head -1 | sed 's/.*version: *//')
  echo "  $name  v$version"
done

Show the user the list and confirm the update is complete.

If backups were created in Step 2, also show: "Backed up N modified files to .agents/update-backup/<date>/. See Step 5 for restore guidance."