Ops Yolo

Runtime Context

Before YOLO analysis, load:

1. Preferences: cat ${CLAUDE_PLUGIN_DATA_DIR:-$HOME/.claude/plugins/data/ops-ops-marketplace}/preferences.json — read owner, timezone, yolo_enabled, all channel configs
2. Daemon health: cat ${CLAUDE_PLUGIN_DATA_DIR}/daemon-health.json — all services must be healthy for comprehensive analysis
3. Secrets: Resolve ALL keys via env → Doppler → password manager: GITHUB_TOKEN, SENTRY_AUTH_TOKEN, LINEAR_API_KEY, AWS_ACCESS_KEY_ID
4. Ops memories: Load ALL files from ${CLAUDE_PLUGIN_DATA_DIR}/memories/ — contact profiles, preferences, topics, donts. YOLO agents need maximum context.

OPS ► YOLO MODE

Load ops-rules before acting. Public repo (no personal data). Outbound: one draft → one approval → one send. If AskUserQuestion / Workflow are missing, follow Rule 10 in ops-rules (Hermes: numbered options / two-turn Telegram card; delegate_task).

Agent Teams support

If CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 is set, use Agent Teams instead of fire-and-forget subagents for the C-suite analysis (Phase 2). This enables:

• Agents can share findings mid-analysis (CEO discovers a revenue blocker → CFO factors it into ROI)
• You can steer agents if early findings change priorities
• Agents coordinate on the consensus recommendation

Team setup (only when flag is enabled):

TeamCreate("yolo-csuite")
Agent(team_name="yolo-csuite", name="ceo", subagent_type="ops:yolo-ceo", ...)
Agent(team_name="yolo-csuite", name="cto", subagent_type="ops:yolo-cto", ...)
Agent(team_name="yolo-csuite", name="cfo", subagent_type="ops:yolo-cfo", ...)
Agent(team_name="yolo-csuite", name="coo", subagent_type="ops:yolo-coo", ...)

After initial analysis, use SendMessage(to="cto", content="CFO flagged $400/mo in waste — does this change your tech-debt ranking?") or similar to cross-pollinate findings between peer agents. The main /ops:yolo orchestrator (this skill) then reads all four analysis files (ceo-analysis.md, cto-analysis.md, cfo-analysis.md, coo-analysis.md) and synthesizes them into the Hard Truths report. yolo-ceo is a parallel peer, not the synthesizer.

If the flag is NOT set, fall back to standard parallel subagents (fire-and-forget, no mid-task steering).

Phase 0 — MANDATORY: Clear cached state, force fresh-state collection

NEVER reuse a prior YOLO session's reports. Stale C-suite analyses misinform decisions. Before pre-gathering anything, run:

# Archive any prior YOLO sessions (don't delete — keep history) then start fresh.
mkdir -p ~/.claude/yolo-archive
for d in /tmp/yolo-*/; do
  [ -d "$d" ] || continue
  ts=$(basename "$d")
  mv "$d" "$HOME/.claude/yolo-archive/${ts}-$(date +%s)" 2>/dev/null || true
done

# Allocate this run's session dir.
SESSION_ID="$(date +%Y%m%d-%H%M%S)-$$"
SESSION_DIR="/tmp/yolo-${SESSION_ID}"
mkdir -p "$SESSION_DIR"
echo "$SESSION_DIR" > /tmp/yolo-current-session
echo "YOLO session: $SESSION_DIR (prior reports archived to ~/.claude/yolo-archive/)"

# Drop any cached pre-gather artifacts so Phase 1 hits live sources.
rm -f /tmp/ops-marketing-dash.cache /tmp/ops-external.cache /tmp/ops-infra.cache 2>/dev/null || true

Each C-suite agent prompt MUST include this line verbatim:

"Use ONLY the pre-gathered data block in this prompt. Do NOT read prior /tmp/yolo-*/ files, do NOT reference cached reports, do NOT cite earlier session findings. This is a fresh-state run; report what is true RIGHT NOW."

Phase 1 — Pre-gather ALL data (live, never cached)

Run all of these simultaneously:

${CLAUDE_PLUGIN_ROOT}/bin/ops-infra 2>/dev/null || echo '{}'

${CLAUDE_PLUGIN_ROOT}/bin/ops-git 2>/dev/null || echo '[]'

${CLAUDE_PLUGIN_ROOT}/bin/ops-prs 2>/dev/null || echo '[]'

${CLAUDE_PLUGIN_ROOT}/bin/ops-ci 2>/dev/null || echo '[]'

${CLAUDE_PLUGIN_ROOT}/bin/ops-unread 2>/dev/null || echo '{}'

${CLAUDE_PLUGIN_ROOT}/scripts/aws-usage-cost.sh snapshot 2>/dev/null || echo '{}'

cat "${CLAUDE_PLUGIN_ROOT}/scripts/registry.json" 2>/dev/null || echo '{}'

${CLAUDE_PLUGIN_ROOT}/bin/ops-external 2>/dev/null || echo '[]'

${CLAUDE_PLUGIN_ROOT}/bin/ops-marketing-dash 2>/dev/null || echo '{}'

${CLAUDE_PLUGIN_ROOT}/bin/ops-gsd-states 2>/dev/null || true

Competitor state: Each C-suite agent sources scripts/lib/competitor/context.sh and calls competitor_vertical_slice <role> internally. The orchestrator does NOT need to pre-fetch competitor data — it is loaded on-demand per agent with role-specific filtering.

Phase 2 — Spawn 4 C-suite agents in parallel

Spawn these 4 agents simultaneously using all pre-gathered data as context. Each writes their analysis to a file in /tmp/yolo-[session]/:

Agent 1 — CEO (Strategic)

Uses agents/yolo-ceo.md. Writes /tmp/yolo-[session]/ceo-analysis.md.

• What's the #1 thing blocking growth right now?
• Are we building the right things?
• Where are we wasting time vs. creating value?
• What would you tell an investor today, unfiltered?

Agent 2 — CTO (Technical)

Uses agents/yolo-cto.md. Writes /tmp/yolo-[session]/cto-analysis.md.

• What's the worst technical debt that will bite us?
• Which services are time-bombs?
• Is the team/architecture set up to scale?
• What corners were cut that need fixing now?

Agent 3 — CFO (Financial)

Uses agents/yolo-cfo.md. Writes /tmp/yolo-[session]/cfo-analysis.md.

• Actual burn rate vs. runway
• Which AWS services are waste?
• When do we hit zero if nothing changes?
• What's the ROI on current work?
• If home_automation is configured in $PREFS_PATH, check energy spend trend via /ops:ops-home status — is the kWh/day curve trending up materially? Flag anomalous draw.

Agent 4 — COO (Operations)

Uses agents/yolo-coo.md. Writes /tmp/yolo-[session]/coo-analysis.md.

• What's falling through the cracks right now?
• Which processes are broken?
• What's the top execution risk this week?
• What should be automated that isn't?
• If home_automation is configured in $PREFS_PATH, include home-automation health in operations checks via /ops:ops-home status: are smoke/water alarms armed? Energy anomalies? Devices offline? Flow failures?

Phase 3 — Hard Truths Report (orchestrator synthesis)

This skill (the main orchestrator) is the synthesizer — NOT yolo-ceo. After all 4 parallel agents complete and have written their analysis files to /tmp/yolo-[session]/{ceo,cto,cfo,coo}-analysis.md, read all four files here in the main context and synthesize them into a unified report.

Phase 3a — Consolidated master executive report (MANDATORY)

Every YOLO run MUST produce one consolidated master executive report — no exceptions. This is the canonical output of the analysis; the four per-officer files are supporting detail. Do this BEFORE rendering the on-screen Hard Truths block:

1. Read all four /tmp/yolo-[session]/{ceo,cto,cfo,coo}-analysis.md files in the main context.
2. Synthesize them into a single executive-summary.md and write it to /tmp/yolo-[session]/executive-summary.md. Never skip this write, even if one officer file is missing — note any missing officer inline and consolidate the rest.
3. The master report MUST contain, in this order:
   • Title + timestamp — YOLO Executive Summary — [date].
   • TL;DR — 2–3 sentences a busy founder can read in 10 seconds.
   • Consensus #1 action — the single most important thing today, with the officers who back it.
   • Per-officer hard truths — the 1–2 sharpest truths from each of CEO / CTO / CFO / COO, summarized (not pasted verbatim).
   • Cross-cutting themes — issues ≥2 officers raised independently (these are the highest-signal items).
   • Prioritized action list — ranked, each tagged [auto] (YOLO can do it) or ⚠️ REQUIRES CONFIRMATION (needs human sign-off), with the source officer(s).
   • What's healthy — 1–2 lines so the report isn't all alarm.
   • Source files — relative links to the four officer files.
4. The on-screen Hard Truths block below is the SHORT view of this same report — keep the two consistent (same consensus, same #1 action).

Render the on-screen report:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 YOLO ► HARD TRUTHS REPORT — [date]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

 CEO: [1-2 brutal strategic truths]

 CTO: [1-2 brutal technical truths]

 CFO: [1-2 brutal financial truths]

 COO: [1-2 brutal operational truths]

──────────────────────────────────────────────────────
 CONSENSUS: The #1 thing that matters today is:
 [single most important action, no sugar-coating]
──────────────────────────────────────────────────────

 Master executive report (consolidated):
 /tmp/yolo-[session]/executive-summary.md

 Supporting per-officer analyses:
 /tmp/yolo-[session]/ceo-analysis.md
 /tmp/yolo-[session]/cto-analysis.md
 /tmp/yolo-[session]/cfo-analysis.md
 /tmp/yolo-[session]/coo-analysis.md

──────────────────────────────────────────────────────
 Type YOLO to hand over the controls.
 I'll run your business autonomously for the next day.
 This means: closing inbox, merging ready PRs,
 fixing fires, advancing GSD phases, triaging issues.

 Or pick an analysis to read:
──────────────────────────────────────────────────────

Use batched AskUserQuestion calls (max 4 options each):

AskUserQuestion call 1:

  [Read master executive report]
  [Read a single officer analysis]
  [Execute top recommendation now]
  [More...]

AskUserQuestion call 2 (only if "More..."):

  [Type YOLO to go autonomous]
  [Stop here]

If the user picks "Read a single officer analysis", present the four officer files (CEO / CTO / CFO / COO) in a follow-up call. The master executive report is the default/recommended read — it already consolidates all four.