agentleFS
metratonmetraton/gaia/skills/gmail-triage/SKILL.md

gmail-triage

Skill4 starsChanged 4 months agoMITView on GitHub

Use when the user wants to clean up, organize, or triage their Gmail inbox interactively -- "ordena el mail", "hagamos triage del inbox", "limpiemos el correo", "let's triage my inbox", "clean up my Gmail".

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before presenting ANY labeled email, check the thread: message count, who sent last, when.

    line 23
  2. Distinctive:

    After each batch: "Moví X a trash, Y a action.

    line 97
  3. Very specific:

    Use --cron '<five fields>' (local time) instead of --every for a fixed clock time, and --workspace to scope it to one workspace.

    line 127
  4. Distinctive:

    When the routine comes due:

    line 129
  5. Distinctive:

    If the user accepts, it runs interactively with them, as "Check My Mail" describes.

    line 131
  6. Very specific:

    When the user says it is done, gaia notifications ack <id> moves the routine to its next occurrence.

    line 132
  7. Very specific:

    If the user says not now, gaia notifications snooze <id> --for 2h (or --until <local time>) hides it until later.

    line 133

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Gmail Triage

Interactive GTD-inspired state machine for Gmail. Gaia analyzes threads, proposes transitions. User decides. Gaia executes. This is the PROCESS layer; it depends on gmail-policy for every rule -- label definitions, security tiers, the interpretation of a review trigger (Intent Classification), and the Autonomous Action Boundary that decides what may run without approval. That dependency holds in every mode: a security rule cannot live only in the process that obeys it. Triage always runs with the user present; to make it recur, see "Making It a Routine".

State Labels

Five _gaia/* labels (defined in gmail-policy) — three durable active states, plus staging and soft-delete:

  • _gaia/action — user must act
  • _gaia/waiting — user acted, awaiting reply
  • _gaia/someday — interesting, no urgency
  • _gaia/pending — staging (triage backlog)
  • _gaia/trash — soft delete; never truly deleted

No _gaia/* label = processed/done.

Thread-Awareness Rule

Before presenting ANY labeled email, check the thread: message count, who sent last, when. This determines framing:

  • "necesitas responder" (user is last)
  • "esperando desde [date]" (user replied, waiting on them)
  • "sin actividad hace 2 semanas — ¿hacer seguimiento?" (stale waiting)

These two sections are the state-machine expression of the Autonomous Action Boundary in gmail-policy — the boundary is the rule, these are its transitions.

Automatic Transitions (no confirmation needed)

  • User replies to an action thread → move to waiting
  • New message arrives in a waiting thread → move to action

Transitions Requiring Confirmation

  • Anything → trash or someday
  • Clearing any label (marking done)
  • someday → action

Modes

Modes 1–5 open with a state summary before their specific work: "Antes de empezar: N en action, N en waiting, N en someday." Flag action items stale >3 days.

0. Check ("chequea mi mail" / "¿algo nuevo?")

  1. Review _gaia/action — present each item with thread framing. Did user already reply? Auto-apply → waiting (mechanical, reversible — no confirmation).
  2. Review _gaia/waiting — did the other party respond? Auto-apply → action (mechanical, reversible — no confirmation). Stale >1 week → flag.
  3. Review _gaia/someday — count only: "tienes 5 en someday." Detail only if asked.
  4. Scan inbox for new signal — Financial (large amounts, bills, due dates), personal/important (housing, legal, health), expected reply arrived → propose action. Interesting, no urgency → propose someday.
  5. Summarize — overall inbox state in 2-3 sentences.

1. Full Triage ("organicemos el correo")

Scan inbox, group by sender/category, report counts. Present top groups. User decides per group → trash/action/someday/content-label. Report progress: "Procesamos 500 de 2000. ¿Seguimos?"

2. Quick Cleanup ("limpiemos algo rápido")

Pick easiest batch (highest volume, most repetitive). "340 promos de retail. ¿Las mando a trash?" One confirmation = hundreds processed. Target: under 2 minutes.

3. Post-Vacation ("acumulé mucho")

Move unprocessed to _gaia/pending. Report: "847 correos: 600 promos, 120 banco, 80 LinkedIn, 47 otros." Work categories in follow-up modes.

4. Review ("¿qué tengo pendiente?")

Dedicated state review — all three active labels:

  • _gaia/action — stale >3 days? move to waiting/someday/done?
  • _gaia/waiting — any responses arrived? stale >1 week?
  • _gaia/someday — weekly review: promote to action? trash any?

5. Promo Analysis ("analiza las promos")

Group by sender, identify patterns. Flag genuinely interesting vs noise. Recommend bulk trash for repetitive senders.

6. Category Sweep ("barre las promociones" / "limpia por categoría" / "¿cuánto tengo en cada categoría?")

Triage by Gmail's native category system labels rather than by sender. The system labels CATEGORY_PROMOTIONS, CATEGORY_SOCIAL, CATEGORY_UPDATES, CATEGORY_FORUMS, and CATEGORY_PERSONAL already partition the inbox — use them as ready-made buckets for high-volume cleanup.

  1. Count per category — get the volume of each category without listing bodies (see "Volume Counting" in gmail-policy/reference.md). Report: "Promotions: 1240, Social: 380, Updates: 610, Forums: 45."
  2. Sweep the highest-volume, lowest-value category first — usually CATEGORY_PROMOTIONS. Propose one bulk transition for the whole bucket: "¿Mando las 1240 de Promotions a trash?"
  3. One confirmation per category — on approval, apply the label change in bulk via batchModify (see "Bulk Label Operations" in gmail-policy/reference.md), chunking to the API's per-call id limit.
  4. Shield protected and transactional senders out of the sweep query first — banks, SII, health, notaries must never be selected, and mixed senders like Ticketplus (promos + real tickets from one address) need the keyword guard. Apply the "Sweep Shield" mechanism (heuristic in gmail-policy Protection Lists, concrete -from: / keyword-guard queries in gmail-policy/reference.md) to the SAME query used for counting and for paging ids, so the count reported and the ids moved are protected identically.
  5. Never sweep CATEGORY_PERSONAL blindly — it holds real correspondence. Sweep it only sender-by-sender, not as a bucket.
  6. Report per category — "Promotions: 1240 → trash. Social: dejé las 380, ¿las revisamos por remitente?"

Category Sweep is the fastest path for post-vacation or long-neglected inboxes; pair it with Mode 3 (Post-Vacation) staging when volume is very high.

Presentation Format

Group by sender/topic. Show count + sample subject. Flag unusual items ("movimiento de $50K en Bci"). Propose action per group. Max 5-7 groups per interaction.

Batch Rules

  • batchModify accepts up to 1000 message ids per call; messages list returns up to 500 per page (paginate for more). Chunk large sweeps to 1000 ids per batchModify. Always confirm before moving: state count and destination.
  • After each batch: "Moví X a trash, Y a action. Z restantes."
  • On "todo trash": double-check — "¿Seguro? Son N correos de [sender]."

Check My Mail

The process that runs when the user gives a general review trigger — "chequea mi mail", "tengo un mail importante", "¿algo nuevo?". The meaning of that trigger — "review with initiative granted", not a bare listing — is defined in gmail-policy (Intent Classification Table); this section is only the PROCESS that meaning invokes. Mode 0 above is the interactive step-list; this section is the contract behind it: one shared triage logic, the mode chosen by the caller.

One process, always with the user present. The user typing "chequea mi mail" — or accepting the offer of a due routine (see "Making It a Routine") — gets quick info plus proposals, live. There is no mode that runs triage without them.

Order of "chequea mi mail" (simple — no watermark, no tracking of the last run):

  1. Review the Gaia filters (_gaia/action, _gaia/waiting, _gaia/someday) and report their state.
  2. Read the inbox and corroborate against those filters.

What the user gets:

  • Give quick info of what there is: what sits in action, who has been waiting for a reply for days, new noise (with an offer to sweep it).
  • Analysis with context — this is the point, what makes it useful rather than a bare listing. Connect information: "hay una promo de X, sé que tienes Santander, te sirve." Detect recurring subscriptions and offer a decision (mantener / desuscribir / spam).
  • Move to filters whatever the user approves.

Autonomy. Every mode obeys the Autonomous Action Boundary in gmail-policy: mechanical, reversible filter moves run without approval; destructive or criterion moves (trash, spam, unsubscribe, delete, marking done, sending) are always proposed to the user. The rule is defined once in the policy layer; this process only obeys it.

Multi-account (brief note). The focus is Gmail. The same filter ORDER is reusable for another Gmail account (e.g. a work address) the day it is connected. If another account has no _gaia/* filters, detect it and be able to create them — gws gmail users labels create is T3 (see gmail-policy), so it is proposed to the user, never run on its own.

Making It a Routine

Triage is a conversation, so a routine does not run it: a routine is a recurring notification that brings the user back to it. Nothing executes on its own; the routine comes due the next time Gaia is used after its time. One command creates it, pointing at this skill:

gaia notifications add --kind routine --every 1d --headline 'Triage mail' --skill gmail-triage

Use --cron '<five fields>' (local time) instead of --every for a fixed clock time, and --workspace to scope it to one workspace; without it the routine is global.

When the routine comes due:

  1. The orchestrator offers to do the triage now.
  2. If the user accepts, it runs interactively with them, as "Check My Mail" describes.
  3. When the user says it is done, gaia notifications ack <id> moves the routine to its next occurrence.
  4. If the user says not now, gaia notifications snooze <id> --for 2h (or --until <local time>) hides it until later.

gaia notifications cancel <id> ends the routine for good.

Anti-Patterns

  • Listing individual emails when hundreds exist — group first, detail on request.
  • Moving without explicit confirmation — removeLabelIds changes visibility with no undo.
  • Auto-processing _gaia/trash — it is the user's safety net, not Gaia's to manage.
  • Assuming promos are trash — some are genuinely interesting. Always ask.
  • Skipping thread check before presenting — framing without thread state misleads the user.
  • More than 5-7 groups per round — decision fatigue kills triage momentum.

Related Skills

  • gmail-policy — security rules, label definitions, operation tiers
  • gws-setup — CLI installation and authentication

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.