Skill: Aggregate Week

When to use

First step of the cockpit cycle. Reads 7 days of source-agent outputs and journals, produces a structured intermediate aggregate that the rendering skill consumes.

Input

Per AGENT.md input contract — 7-30 days of agent outputs from data-bi, performance-monitor, incident-commander, tech-budget-finops, competitive-intel, both LinkedIn agents, investor-relations, market-research, and the shared journal.

Output

Internal structured aggregate (not user-facing). Used by render-cockpit.md.

week_iso: "2026-W22"
date_range_label: "May 19 - May 25"
quarter: "Q2"
quarter_week: 9

shipped:
  - source: "linkedin-bilal-cpo"
    description: "4 new posts written for queue 2026-05-26 (morning-brief-ship, doppler-discipline, self-hosted-runner, markdown-only-agents)"
    evidence: "outputs/queue_2026-05-26/"
  - source: "competitive-intel"
    description: "Q2 positioning brief published (5-competitor matrix, 5 plays)"
    evidence: "outputs/2026-05-25_brief_q2-positioning.md"

slipped:
  - source: "morning-brief"
    description: "Workflow scheduled but no confirmed delivery yet — waiting on Doppler secret setup"
    blocker: "TELEGRAM_BOT_TOKEN / TELEGRAM_CHAT_ID / GOOGLE_OAUTH_* secrets in Doppler"
    age_days: 1

metrics_wow:
  - name: "Daily LLM spend (avg)"
    this_week: "$0.04"
    last_week: "$0.03"
    delta: "+$0.01"
    note: "expected — new founder-cockpit weekly tier added"
  - name: "Active SERP position-1 queries"
    this_week: 3
    last_week: 3
    delta: 0
    note: "stable"

content:
  linkedin_bilal:
    shipped_this_week: 0
    queue_size: 7
    next_post: "morning-brief-ship — Mon 5/26 09:30 TRT"
  linkedin_emir:
    shipped_this_week: 0
    queue_size: 7
    next_post: "market-lens-dm-economy — Tue 5/27 09:30 TRT"
  social_media:
    shipped_this_week: 0
    queue_size: "(no output this week)"

pipeline:
  salons:
    new_conversations: 0
    anchor_confirmations: 0
    pipeline_total: 0
  investors:
    active_conversations: 0
    new_intros: 0
    deals_at_stage:
      first_meeting: 0
      term_sheet_review: 0
      committed: 0
  hires:
    active_candidates: 0

decisions_needed:
  - id: "DN-2026W22-1"
    description: "Telegram bot setup for morning-brief — needs personal chat_id confirmation before workflow can deliver"
    deadline: "Mon 5/26 06:00 TRT (before first scheduled run)"
    estimated_effort: "5 min"
  - id: "DN-2026W22-2"
    description: "First LinkedIn post (morning-brief-ship) ready for review — approve or revise"
    deadline: "Mon 5/26 09:30 TRT"
    estimated_effort: "10 min"

next_week_bets:
  - "Verify morning-brief end-to-end delivery + ship Mon LinkedIn post"
  - "Approve + schedule 8 LinkedIn posts via Publora or manual"
  - "Send first 2 cold investor emails from outreach pack (TR angel + TR family office)"

risks_flagged:
  - source: "founder-cockpit RULES.md"
    severity: low
    description: "morning-brief delivery unverified — risk of silent failure first day"

Procedure

1. Date math. Compute Sunday 00:00 TRT this week → 23:59 TRT this week as window. Convert to UTC for filesystem mtime filter.
2. Read shipped. For each source agent, scan outputs/ for files with mtime in window. Each becomes a shipped item.
3. Read slipped. For each source agent with a HEARTBEAT.md cadence shorter than weekly (daily, weekly), check if expected outputs exist for this week. Missing = slipped.
4. Pull metrics. From data-bi outputs (latest in window), extract topline KPIs. Compare to prior week. Compute delta. If source missing, mark (stale).
5. Content + pipeline. Read LinkedIn agent queues, social-media-manager queues, sales-bd outputs, investor-relations outputs. Count + summarize.
6. Decisions. Scan all journal entries for Risk: lines and (needs founder decision) markers. Aggregate.
7. Next week. Read each source agent's most recent (planned next cycle) or (next cadence) markers. Synthesize top 3.
8. Output the YAML above. Pass to render-cockpit.

Rules

• If any source returns no output for the window, log it as (no output this week) in the aggregate, do NOT skip silently
• If >2 critical sources (data-bi, performance-monitor, morning-brief) are missing >3 days this week, set degraded_critical: true in the aggregate — render-cockpit will switch to alert mode
• Time math is TRT (Europe/Istanbul, UTC+3) — never UTC in the user-facing output
• All quoted descriptions must be verbatim from source agent outputs — never paraphrase claims that have numbers in them

Skill: Render Cockpit

When to use

Final step of the cockpit cycle. Consumes the aggregate YAML from aggregate-week.md and produces the user-facing Telegram body (strict shape) + journal entry.

Input

• The aggregate YAML object produced by aggregate-week.
• MEMORY.md tail (last 200 lines) for format/voice context.

Output Shape (strict — this is what gets pushed to Telegram)

## <ISO timestamp>
**Action:** rendered week N of Q{Y} cockpit covering N decisions and M next-week bets
**Output:** (Telegram push — see body)
**Learned:** <one durable observation from the week, or "nothing new">
**Handoff:** none (leaf agent)
**Risk:** <highest-severity risk from aggregate, or "none">

### 📊 Week {N} of Q{Y} ({{date range}})

**🚀 Shipped this week**
- <source>: <description>
- ...

**🚧 Slipped or blocked**
- <source>: <description> ({{age_days}}d)
- ...

**📈 Metrics (week-over-week)**
- <name>: <this_week> vs <last_week> ({{delta sign + value}})
- ...

**📣 Content & comms**
- LinkedIn Bilal: shipped {{N}} / queue {{M}}, next: {{title + day/time}}
- LinkedIn Emir: shipped {{N}} / queue {{M}}, next: {{title + day/time}}
- Social: {{shipped or "(quiet week)"}}

**🤝 Pipeline**
- Salons: {{N}} new conversations, {{M}} anchor confirmations
- Investors: {{N}} active, {{M}} new intros
- Hires: {{N}} candidates active

**🎯 Decisions needed**
1. [{{id}}] {{description}} — by {{deadline}} ({{effort}})
2. ...

**📅 Next week's top 3 bets**
1. ...
2. ...
3. ...

{{Optional footer: ⚠️ <single critical risk line if any>}}

Procedure

1. Compute header line. Week ISO + quarter + date range from aggregate.
2. Shipped section. Iterate aggregate.shipped[]. Format as bullet per source. If empty, write (quiet week — no new shipped items).
3. Slipped section. Iterate aggregate.slipped[]. Append age days. If empty, write (none flagged).
4. Metrics section. Format each delta with sign (+/-/=). For percentages, prepend %. For currency, prepend $. Always show this_week and last_week side by side.
5. Content section. Render LinkedIn + social lines per template. If queue is (no output this week), write (queue stale — needs source).
6. Pipeline section. Three lines (salons / investors / hires). If counts are all zero, write (quiet week — no new activity).
7. Decisions section. Iterate aggregate.decisions_needed[]. Number 1-N. Cap at 5 — if more, take the 5 highest-priority and write (N more — see outputs/decisions_{YYYY-MM}.md) at end.
8. Next week's bets. Iterate aggregate.next_week_bets[]. Cap at 3.
9. Risk footer. If aggregate.risks_flagged[] has any item with severity high or critical, add the line at the bottom with ⚠️ prefix.
10. Length check. Total body ≤ 2500 chars. If over, truncate Decisions list first, then Metrics, then Slipped. Always preserve Shipped + Next week's bets + Decisions IDs.
11. Header block. Add the strict-shape header (Action/Output/Learned/Handoff/Risk) ABOVE the ### 📊 line so send_telegram.mjs extractBody() strips it correctly.
12. Write to journal. Path: agents/journal/<UTC-YYYY-MM-DD>_founder-cockpit_autopilot.md. Append (not overwrite) — multiple cockpit cycles in a day are allowed but rare.

Degraded Mode

If aggregate.degraded_critical == true, produce a single-line alert body instead:

### 🚨 Weekly cockpit blocked

{{N}} agents stale, {{M}} fetchers missing. See GitHub Actions for detail.

Decisions still surfaced (if any):
1. ...
2. ...

This still pushes to Telegram — the founder needs to know the system is degraded, not just see silence. Decisions still surface because they're the highest-leverage section.

Rules

• Telegram body ≤ 2500 chars (single message)
• All times in TRT (UTC+3), never UTC in user-facing text
• No markdown tables (Telegram strips them)
• Light emoji per section header only — no emoji storms in body
• Section order is canonical per RULES.md, never reorder
• If a metric is (stale), do not show a delta — just (stale this week)
• Round status is one line, not multi-line — detail lives in investor-relations
• Customer / salon names anonymized per RULES.md