Skill: Cap Table Tracking

Purpose

Track SAFEs, convertible notes, founder equity, and option pool; model dilution under funding scenarios. Read-only on source data; changes only mirror HUMAN + legal-approved actions.

Serves Goals

• Data room completeness (cap table is a mandatory section)
• Pipeline velocity (clean cap table view accelerates DD)

Inputs

• finance-fpa authoritative cap table source
• HUMAN-provided SAFE/note documents in data/imports/
• Previous outputs/YYYY-MM-DD_cap_table.md
• Proposed funding scenario parameters from HUMAN (e.g. $500K at $4M cap)

Process

1. Read the latest cap table source from finance-fpa. Treat as single source of truth.
2. Enumerate instruments: founder common (Anastasia, Bilal), option pool, any prior SAFEs/notes with cap/discount/MFN terms.
3. For each pending instrument in data/imports/, verify HUMAN + legal approval note exists. If not, flag and skip.
4. Produce current ownership snapshot (fully diluted %).
5. Model 1-3 dilution scenarios: e.g. current round at $3M cap, $4M cap, $5M cap; and post-Series A illustrative round.
6. For each scenario: show pre-money, post-money, per-party dilution, option pool impact.
7. Write to outputs/YYYY-MM-DD_cap_table.md with current state + scenarios.
8. Log journal entry with summary; flag HUMAN for any discrepancy between finance-fpa source and prior snapshot.
9. Never mutate finance-fpa source. If a change is needed, escalate to HUMAN + legal counsel.

Outputs

• outputs/YYYY-MM-DD_cap_table.md with current ownership and 1-3 scenarios
• Journal entry summarizing delta vs prior snapshot

Quality Bar

• Zero divergence from finance-fpa cap table source
• Every scenario shows math: pre-money, round size, new shares, option pool top-up, per-party dilution
• Fully-diluted percentages sum to 100.0% +/- 0.1%
• HUMAN + legal approval referenced for any instrument included

Tools

• Read tool for finance-fpa source
• scripts/dilution_calc.sh (future) for scenario math validation

Integration

• Consumes finance-fpa cap table source (authoritative)
• Output consumed by DATA_ROOM_MAINTENANCE (Company section) and PITCH_DECK_GENERATION (ask slide context)
• Any cap table question in FUNDRAISING_PIPELINE routes here first

Skill: Cold Outreach

Purpose

Draft segmented investor cold emails, forwardable warm-intro blurbs, and follow-ups, ready for founder review. Complements FUNDRAISING_PIPELINE (which tracks the funnel) by feeding the top of it.

Serves Goals

• Pipeline velocity (more qualified intros into the Intro stage)
• Consistent, on-message founder voice across many sends

Inputs

• knowledge/STRATEGY.md, knowledge/BRAND.md (positioning, voice)
• competitive-intel/outputs/ latest brief + battlecard (qualitative wedges, pre-cleared)
• market-research/outputs/ latest sizing (market-context numbers, sourced)
• data-bi + finance-fpa outputs (traction metrics, REQUIRED before any send)
• A target list with one why_them reason per investor (from HUMAN)

Process

1. Segment the target list. Default segments: (A) consumer/beauty-tech VC, (B) regional Turkey/CEE-MENA fund, (C) marketplace-thesis generalist, (D) operator/angel, (E) accelerator.
2. For each segment, write a <150-word email: one specific hook, the local wedge, one traction line, one clear ask (deck or 20 min).
3. Use qualitative positioning freely (no-acquisition-tax, Instagram-DM incumbent, AI-first beauty depth). These are pre-cleared via the intel refresh.
4. Put EVERY traction metric in [[ ]] placeholders sourced from data-bi/finance-fpa. Never state a number from memory. Revenue/GMV/raise need clearance.
5. Produce a forwardable warm-intro blurb (<80 words) addressed to the mutual connection, not the investor.
6. Produce a 5-7 day follow-up with a fresh proof point slot.
7. Provide TR variants for clearly Turkish-first recipients.
8. Write to outputs/YYYY-MM-DD_cold_outreach_pack.md. Flag for HUMAN approval before any send. Log to journal.

Outputs

• outputs/YYYY-MM-DD_cold_outreach_pack.md with: a "metrics to fill" table, 5 segment templates, accelerator blurb, warm-intro blurb, follow-up, sender notes.

Quality Bar

• lowercase glossgo, no emojis, English primary (TR variants where marked)
• No traction number outside a [[ ]] placeholder until filled from a real source
• Every email has a single specific ask and a personalized {{why_them}}
• Competitor scan: name a competitor only in internal notes, never in the email body (mirror Emir's LinkedIn rule)
• Outward-facing, so nothing sends without HUMAN approval (per RULES)

Reference

See outputs/2026-06-03_cold_outreach_pack.md for a complete worked example.

Skill: Data Room Maintenance

Purpose

Keep the structured data room current across company docs, financials, KPIs, legal, product, and team sections so DD requests can be fulfilled without delay.

Serves Goals

• Data room completeness (score >8/10)
• Pipeline velocity (fast DD response shortens intro-to-term-sheet cycle)

Inputs

• finance-fpa: latest projections, cap table source, burn and runway
• data-bi: KPI snapshots
• kvkk-compliance: KVKK posture, DPA templates, compliance audits
• product-roadmap: product specs, architecture overview
• tech-budget-finops: infra cost breakdown, Cloudflare/Supabase/iyzico line items
• Legal docs provided by HUMAN (incorporation, IP assignments, contracts)
• Previous outputs/YYYY-MM-DD_data_room_index.md

Process

1. Walk the canonical data room sections: Company (incorporation, bylaws, cap table), Financials (projections, burn, runway, bank statements), KPIs, Legal (IP, contracts, KVKK), Product (specs, roadmap, architecture), Team (bios, org chart, option pool).
2. For each section, check the latest artifact date against a 90-day freshness threshold.
3. Mark each item: present+fresh, present+stale, missing, blocked on HUMAN.
4. Score completeness 1-10 using: (fresh items / total required items) * 10, rounded.
5. Produce outputs/YYYY-MM-DD_data_room_index.md listing every item with status, path, and owner agent or HUMAN.
6. For missing or stale items, emit specific action lines (e.g. "Request updated KVKK audit from kvkk-compliance").
7. Log journal entry with score and top 3 gaps.
8. Run quarterly deep audit: also verify external links (Notion, Drive) are reachable.

Outputs

• outputs/YYYY-MM-DD_data_room_index.md with all sections, item status, completeness score
• Journal entry with score delta vs prior audit and top gaps

Quality Bar

• Completeness score >8/10 sustained
• Zero missing items in Company and Financials sections at any time
• Every item has an owner (agent slug or HUMAN) and a path
• Legal and KVKK items flagged for HUMAN review even when present

Tools

• Read tool for upstream outputs
• scripts/data_room_check.sh (future) to validate external links

Integration

• Consumes finance-fpa, data-bi, kvkk-compliance, product-roadmap, tech-budget-finops outputs
• Gaps flagged here escalate to HUMAN or the owning agent via journal
• Completeness score reported in monthly INVESTOR_UPDATE metrics block

Skill: Fundraising Pipeline

Purpose

Maintain the investor CRM through stages (intro, meeting, DD, term sheet, close) with weekly updates and next-step alerts for each prospect.

Serves Goals

• Pipeline velocity (intro to term sheet <45 days)
• Monthly update cadence (pipeline data feeds Asks section)

Inputs

• HUMAN-provided investor CRM export in data/imports/ (CSV or markdown)
• Founder meeting notes in data/imports/ or journal/
• Previous outputs/YYYY-MM-DD_pipeline_status.md
• Journal entries tagged investor-relations

Process

1. Load latest CRM export. If >7 days old, request fresh export from HUMAN.
2. Normalize each investor record: name, firm, stage, last contact date, next step, owner (usually HUMAN founder).
3. Classify stages: Intro, First Meeting, Second Meeting, DD, Term Sheet, Closed, Passed.
4. Compute cycle time per stage and intro-to-term-sheet average across active prospects.
5. Flag stalled prospects (no movement in 14 days) with a recommended next action (founder follow-up, deck share, DD response).
6. Flag high-velocity prospects (multiple stage jumps in a week) for prioritization.
7. Cross-reference with DATA_ROOM_MAINTENANCE: if an investor is in DD and data room has gaps, escalate immediately.
8. Produce outputs/YYYY-MM-DD_pipeline_status.md with stage table, stalled list, action list, cycle time metric.
9. Log journal entry with top 3 actions for HUMAN this week.

Outputs

• outputs/YYYY-MM-DD_pipeline_status.md with stage table, stalled prospects, next actions, cycle time metric
• Journal entry with HUMAN action list

Quality Bar

• Every active prospect has a stage, last contact date, and next step
• Stalled prospects (>14 days no movement) always flagged with a recommended action
• Intro-to-term-sheet cycle time computed and trended weekly
• No PII shared outside data room; investor names internal only
• English only, no emojis

Tools

• Read tool for CRM export
• scripts/pipeline_cycle.sh (future) to compute cycle time from CSV

Integration

• Feeds Asks section of INVESTOR_UPDATE (who to intro)
• Escalates to DATA_ROOM_MAINTENANCE when DD-stage gaps are detected
• References deck version from PITCH_DECK_GENERATION per-investor send log
• Surfaces HUMAN-only actions (negotiations, term sheet response) for weekly founder review

Skill: Investor Update

Purpose

Draft the monthly investor update letter covering KPIs, wins, misses, and asks, ready for founder review and send by day 5.

Serves Goals

• Monthly update cadence (100% sent by day 5)
• Pipeline velocity (consistent updates keep warm investors engaged)

Inputs

• finance-fpa latest month close: revenue, GMV, burn, runway
• data-bi month-end KPIs: salons onboarded, bookings, retention, NPS
• product-roadmap: shipped features, next month roadmap
• partnerships and sales-bd recent wins
• Journal entries since last update: decisions, hires, incidents
• Previous month's investor update in outputs/ for continuity

Process

1. Confirm today is between day 1 and day 5 of the month. If not, do not run unless HUMAN requests off-cycle update.
2. Pull prior update; identify asks that were fulfilled, dropped, or still open.
3. Compose sections in fixed order: Highlights, Metrics, Product, Team, Asks.
4. Highlights: 3-5 bullets. Wins AND misses. No marketing prose.
5. Metrics: table with this month vs last month vs target (from AGENT.md KPIs, finance-fpa, data-bi).
6. Product: shipped this month, shipping next month. Reference product-roadmap slugs.
7. Team: hires, departures, open roles.
8. Asks: 2-4 specific, actionable requests (intros to named investors, hiring referrals, customer intros).
9. Write to outputs/YYYY-MM-DD_investor_update.md. Cross-check numbers against finance-fpa and data-bi outputs.
10. Log journal entry, flag for HUMAN approval before send.

Outputs

• outputs/YYYY-MM-DD_investor_update.md with five fixed sections
• Journal entry with approval status and send date once HUMAN confirms

Quality Bar

• Sent by day 5 of every month; cadence is the KPI
• Both wins and misses included; never hide a miss
• Every metric traces to finance-fpa or data-bi output
• Asks are specific (named targets or roles), not generic ("intros welcome")
• English only, no emojis, lowercase "glossgo"

Tools

• Read tool for upstream agent outputs
• scripts/metrics_snapshot.sh (future) to auto-pull month-end KPIs

Integration

• Consumes finance-fpa, data-bi, product-roadmap, partnerships, sales-bd outputs
• Asks section seeds FUNDRAISING_PIPELINE next-step alerts (intros to chase)
• Metrics block shared with PITCH_DECK_GENERATION traction slide

Skill: Pitch Deck Generation

Purpose

Assemble and update the glossgo seed pitch deck by composing slides from finance-fpa, competitive-intel, and product-roadmap outputs into a single canonical narrative.

Serves Goals

• Pitch deck freshness (days since last update <30)
• Deck quality (founder-confirmed ready-to-send rating >4/5)

Inputs

• Latest finance-fpa output: projections, unit economics, burn, runway, Year 1 and Year 5 targets
• Latest competitive-intel brief: market landscape, differentiation
• Latest product-roadmap milestones and release timeline
• Latest data-bi KPIs: salons, bookings, GMV, retention
• knowledge/STRATEGY.md for narrative frame
• Previous deck in outputs/ for diff reference

Process

1. Pull the five most recent upstream outputs (finance-fpa, competitive-intel, product-roadmap, data-bi, market-research). Abort if any is older than 30 days; escalate to HUMAN.
2. Verify the ask is current: $500K-$750K seed at $3M-$5M pre-money cap, SAFE preferred, 18-month runway.
3. Draft slides in canonical order: company overview, problem, solution, product, market, business model, traction, team, ask.
4. Cross-check every number against finance-fpa and data-bi outputs verbatim. Flag any mismatch; do not invent values.
5. Keep brand name lowercase: glossgo. No emojis. English only.
6. Run length check: each slide <60 words except traction and team.
7. Write to outputs/YYYY-MM-DD_pitch_deck_v<major>.<minor>.md. Bump minor on content edit, major on narrative restructure.
8. Log journal entry with version, changes summary, and unresolved review items. Flag for HUMAN approval before send.

Outputs

• outputs/YYYY-MM-DD_pitch_deck_v<major>.<minor>.md containing all nine slides in markdown
• Journal entry noting version bump and open review items

Quality Bar

• Every numeric claim traces to a finance-fpa or data-bi output with a citation in a comment
• Nine canonical slides present in order; no extra slides without HUMAN approval
• Founder reviews and rates >=4/5 "ready-to-send" before external use
• Zero Turkish characters, zero emojis, zero marketing prose
• Valuation/ask block matches current fundraising parameters

Tools

• Read tool for upstream agent outputs under agents/<slug>/outputs/
• scripts/deck_diff.sh (future) to compare current vs previous deck

Integration

• Consumes finance-fpa, competitive-intel, product-roadmap, data-bi, market-research outputs
• Feeds into INVESTOR_UPDATE (shared numbers block) and DATA_ROOM_MAINTENANCE (deck is a data room asset)
• Version referenced by FUNDRAISING_PIPELINE when logging which deck was sent to which investor