Ops Deploy Fix

Subcommand: test

Synthetic dry-run through the pipeline. No real fix dispatch — no agent runs. Confirms wiring: hooks fire → monitor classifies → notify channel pings → state files written → would-dispatch path logged.

Steps:

1. Pre-flight check:

   PLUGIN_ROOT="${CLAUDE_PLUGIN_ROOT:-}"
   [ -z "$PLUGIN_ROOT" ] && PLUGIN_ROOT=$(find ~/.claude -type d -name claude-ops 2>/dev/null | head -1)
   COMMON="$PLUGIN_ROOT/scripts/lib/deploy-fix-common.sh"
   [ -f "$COMMON" ] || { echo "✗ deploy-fix-common.sh missing — re-install the plugin"; exit 1; }

2. Ask:

   What should the synthetic failure look like?
     [Build failure (npm run build:* exit 1)]
     [Deploy workflow failure (gh actions check_run)]
     [Health check failure (curl /health 503)]
     [Version mismatch (served SHA != merged SHA)]

3. Generate a fake event payload matching the chosen kind. Set env var OPS_DEPLOY_FIX_DRY_RUN=1 and invoke the relevant trigger script:

   export OPS_DEPLOY_FIX_DRY_RUN=1
   export OPS_DEPLOY_FIX_TEST_REPO="your-org/your-repo"   # placeholder per Rule 0
   export OPS_DEPLOY_FIX_TEST_BASE="dev"
   case "$kind" in
     build)   bash "$PLUGIN_ROOT/bin/ops-deploy-fix-build-trigger" --synthetic ;;
     deploy)  bash "$PLUGIN_ROOT/bin/ops-deploy-fix-merge-trigger" --synthetic ;;
     health)  bash "$PLUGIN_ROOT/scripts/ops-deploy-monitor.sh" --synthetic-health ;;
     version) bash "$PLUGIN_ROOT/scripts/ops-deploy-monitor.sh" --synthetic-version ;;
   esac

   Trigger scripts must honor OPS_DEPLOY_FIX_DRY_RUN=1 by:

   • Skipping the actual claude -p ... dispatch in dispatch_fix_agent
   • Writing a would-dispatch-<id>-<ts>.log to $LOGS_DIR instead
   • Still incrementing the budget counter (so the test exercises the cap)
   • Logging [DRY-RUN] would notify <channel> instead of firing real notify (no actual outbound messages during test — Rule 6)

   If the trigger scripts don't yet support --synthetic / OPS_DEPLOY_FIX_DRY_RUN, surface that as a known TODO in the output and degrade to "would-have-dispatched" log-only.

4. Print the resulting would-dispatch-*.log path and a one-line classification result (transient? dedup-hit? budget-exhausted? would-dispatch-template=<file>).

5. Offer:

   Test complete. Next?
     [Tail the would-dispatch log]
     [Run another test]
     [Show status dashboard]
     [Done]

Failure modes / hand-offs

• No prefs file → tell the user to run /ops:setup (or /ops:deploy-fix configure) first.
• claude CLI missing (status / test) → instruct via output, do not auto-install.
• jq missing → run brew install jq in background (Rule 4) and retry.
• Stale lock detection → see status flow above, requires Rule-5 confirmation to clear.
• Notify channel mis-set (e.g. discord selected but no webhook) → on status, surface as ⚠️ notify_channel=discord but no webhook URL configured and offer [Reconfigure now].

Files this skill reads / writes

• Read-only: ~/.claude/plugins/data/ops-ops-marketplace/preferences.json, ~/.claude/state/ops-deploy-fix/*, ~/.claude/logs/ops-deploy-fix/*, ${CLAUDE_PLUGIN_ROOT}/scripts/lib/deploy-fix-common.sh, ${CLAUDE_PLUGIN_ROOT}/prompts/{build-fix,deploy-fix}.md
• Write (with explicit confirmation only): lock files in ~/.claude/state/ops-deploy-fix/lock-* (clear on stale)
• Write (via configure subcommand): the prefs file, via the merge pattern in Step 6.5