Inbox

Inbox — Email Triage System

Self-contained email triage skill. Handles setup and recurring triage in a single command. Detects its own state and routes accordingly — no need to remember which mode to run.

State Detection (Always First)

Before anything else, check for the KB:

Edge cases:

• User says "set up my inbox" but KB exists → acknowledge, ask: "run triage / update preferences / both?"
• User says "triage my inbox" but no KB → "No knowledge base found. Running setup first, then triage immediately after." → Setup → Triage
• Partial KB (some files missing) → list which files are missing, ask: "run full setup / fill in only the missing sections?"

---

SETUP MODE — Build the Knowledge Base

Runs when no KB exists.

Conduct Discipline

Do NOT generate all files at once. Walk through the 8 sections one at a time. Each section commits its file(s) before moving on. Partial completion (user drops off mid-interview) still produces a usable partial KB.

• One question per turn. Never bundle — even across section boundaries.
• "Why I'm asking" on every question — users answer better when they know what's at stake.
• Forcing format where possible. Multi-choice > open-ended.
• Dependency-ordered. Section N informs Section N+1.

See references/grill_me_section_walk.md for the 8-section discipline detail.

Knowledge Base — 7 Files to Produce

Write to ${WORKSPACE}/Email/:

Stop Condition

~25–31 questions total across 8 sections. Hard ceiling: 35 questions including sub-clarifications. Section 4 (Evaluation Framework) is skipped entirely when Section 1 surfaced no opportunity-email category. After Section 8's confirmation + handoff message, intake is closed — to change preferences, use update mode.

Section 1: The Big Picture

Six grill-me questions, one at a time:

• S1.Q1: "What do you do? Give me your role and business in 1–2 sentences. Why I'm asking: Context shapes what email patterns to expect — a solo creator's inbox looks nothing like an enterprise PM's."
• S1.Q2: "What dominates your inbox? Pick the top 1–2: sales pitches / client work / internal team / newsletters / customer support / financial / other. Why I'm asking: Dominant categories drive the taxonomy."
• S1.Q3: "Rough volume split — e.g., '60% business inquiries, 20% ops, 20% noise'. Why I'm asking: The split tells me where to focus triage effort."
• S1.Q4: "Which email address(es) should triage cover? Why I'm asking: If multiple, I'll set up per-address taxonomies."
• S1.Q5: "Run frequency: once daily / 2x daily / 3x daily / on-demand only? Why I'm asking: Drives the default search window in triage (9h overlap for 2x/day)."
• S1.Q6: "Anyone helping manage email — assistant, VA, team — or solo? Why I'm asking: Persona handling differs for delegated inboxes."

Action: Build mental model. Do NOT write files yet. Note whether opportunity emails are a category (drives S4 skip-logic).

Section 2: Email Categories

Propose 5–7 categories based on Section 1 — pre-recommend a subset:

• New Opportunities
• Active Conversations
• Action Required
• Financial
• Important/Personal
• Informational
• Ignore/Low Priority

Three forcing questions, one at a time:

• S2.Q1: "Here's my proposed taxonomy: [list]. Does this match your inbox reality — yes / mostly / no? Why I'm asking: If 'no', I need to redo the taxonomy before any other section makes sense."
• S2.Q2: "Missing categories? List them. (Skip if none.) Why I'm asking: Missing categories produce uncategorized emails downstream, which hurts triage quality."
• S2.Q3: "Which category takes the MOST time per email? Why I'm asking: That's where draft-reply effort needs to focus most."

Action: Generate email-taxonomy.md with categories, signals, and default actions per category.

Section 3: Reply Style & Voice

Six grill-me questions plus the critical sample request:

• S3.Q1: "Register: formal / casual / in-between? Why I'm asking: Calibrates default voice; we'll refine from samples next."
• S3.Q2: "Three communication pet peeves — phrases you hate, openings you avoid. Why I'm asking: I treat these as forbidden tokens in drafts."
• S3.Q3: "Phrases or sign-offs you always use — list as many as come to mind. Why I'm asking: These are your voice fingerprints."
• S3.Q4: "Different persona for different contexts — e.g., assistant replies as you? Why I'm asking: Persona context changes pronoun + signature handling."
• S3.Q5: "Typical reply length — one-liner / short paragraph / longer? Why I'm asking: Length is the easiest voice signal to get wrong."
• S3.Q6: "Hard rules — never X / always Y? Why I'm asking: Hard rules are enforced as non-negotiable in every draft."

S3.SAMPLES (highest-quality input):

Paste 3–5 real sent emails from your inbox.

Why I'm asking: Self-description of voice is unreliable. Real samples are the best signal — use scripts/voice_sample_analyzer.py to extract patterns.

Action: Generate email-patterns.md. See references/voice_calibration.md for the sample-extraction discipline.

Section 4: Evaluation Framework (Conditional)

Skip if Section 1 surfaced no opportunity emails. Otherwise six grill-me questions:

• S4.Q1: "First thing you check when pitched something — give me your gut filter. Why I'm asking: That's the top of the decision tree."
• S4.Q2: "Three instant deal-breakers. Why I'm asking: These become PASS-auto signals."
• S4.Q3: "Three things that make you immediately interested. Why I'm asking: These become TAKE-IT signals."
• S4.Q4: "Standard pricing / terms — or 'no fixed pricing'? Why I'm asking: If you have a rate card, I'll generate one."
• S4.Q5: "Negotiation posture: firm / flexible / depends on context? Why I'm asking: Drives draft tone on counter-offers."
• S4.Q6: "VIP senders or organizations that always get engagement. Why I'm asking: VIP list bypasses normal PASS filters."

Action: Generate evaluation-framework.md + rate-card.md (if pricing exists).

Section 5: Blocklist & Patterns

• S5.Q1: "Senders or domains to always skip. (Skip if none.) Why I'm asking: Auto-blocklist saves the most time per run."
• S5.Q2: "Patterns in emails you always delete. Why I'm asking: Patterns let triage auto-skip variants without exact-match maintenance."
• S5.Q3: "Specific companies / recruiters / newsletters wasting time. Why I'm asking: These seed the blocklist; triage will add more as you override decisions."

Action: Generate blocklist.md.

Section 6: Current State

• S6.Q1: "Active threads you're tracking — list with one-line context each. (Skip if none.) Why I'm asking: These become tracker entries so triage knows existing context."
• S6.Q2: "Overdue replies? Why I'm asking: Triage flags these as priority every run until resolved."
• S6.Q3: "Time-sensitive items with deadlines? Why I'm asking: Tracker enforces deadlines and surfaces them as overdue at the right time."

Action: Generate tracker.md + create empty triage-log/ directory.

Section 7: Report Preferences

• S7.Q1: "Delivery format: email draft to self / file in workspace / chat summary only. Why I'm asking: The triage report goes here every run."
• S7.Q2: "Detail level: 30-second scan / detailed breakdown / both. Why I'm asking: Affects report length."
• S7.Q3: "Anything always shown first — e.g., overdue payments, VIP messages? Why I'm asking: Custom top-of-report rules surface what you care about above standard sections."

Action: Append report preferences to email-taxonomy.md.

Section 8: Confirmation & Handoff

List every file created with a one-sentence summary. Run scripts/kb_validator.py --workspace ${WORKSPACE} to confirm the 7-file contract is satisfied.

Then ask:

Your triage system is ready. Run triage now?

1. Yes — process my inbox immediately
2. No — I'll invoke it separately

If yes → transition directly to Triage Mode below.

Privacy Boundary

Never persist passwords, full account numbers, SSNs, or other sensitive credentials in KB files. If volunteered during the interview, acknowledge but don't store — write [stored separately by user] in the relevant file.

---

TRIAGE MODE — Recurring Email Triage

Runs when KB exists. Light-intake by design.

DRAFTS ONLY — Never Send

This skill creates drafts. It NEVER sends.

Non-negotiable. The scripts/draft_safety_validator.py enforces it post-run. Any send-shaped tool call in the action log fails validation. See references/drafts_only_safety.md.

Step 0: Grill-Me Intake (0–2 Optional Override Questions)

Q1 (only when on-demand run is outside normal cadence):

Override the default 9-hour search window? yes (specify hours) / no.

Why I'm asking: If you're running outside your normal cadence, you may want a wider or narrower window.

Q2 (only when user invokes with category-skip intent):

Skip any categories this run? E.g., "skip newsletters".

Why I'm asking: Sometimes you just want to scan opportunities or clear active threads.

Max 2 questions. Default invocations skip both and run with KB-default preferences.

Step 1: Determine Search Window

Use scripts/search_window_calculator.py --cadence <CADENCE> --now <ISO>: