inbox-triage
by alirezarezvaniinbox-triage runs recurring or on-demand email triage from an inbox-setup knowledge base. It classifies recent mail, researches senders, recommends actions, drafts replies without sending, logs results, and updates `${WORKSPACE}/Email/` using helper scripts for KB validation, search windows, and draft safety.
This skill scores 84/100, making it a solid listing candidate for directory users who want a recurring or on-demand email triage workflow. The repository evidence shows enough trigger guidance, workflow detail, safety discipline, and supporting scripts for an agent to execute with less guesswork than a generic prompt, but users should understand that it depends on the paired `inbox-setup` knowledge base and lacks local install instructions.
- Highly triggerable: the frontmatter lists many natural invocation phrases such as “triage my inbox,” “check my email,” and “process my inbox,” and states the recurring/on-demand use case clearly.
- Operationally substantive: the skill defines a KB contract, fail-fast behavior, recent-email search window logic, classification, sender research, recommendations, draft generation, reporting, and KB updates.
- Good safety and automation support: references emphasize “drafts only — never send,” and `draft_safety_validator.py`, `kb_reader.py`, and `search_window_calculator.py` provide deterministic checks/utilities.
- Requires the paired `inbox-setup` skill and its exact 7-file `${WORKSPACE}/Email/` knowledge base before it can operate; this is a real dependency, not a standalone inbox workflow.
- No install command or README is present in the skill directory, so directory users may need repository-level installation context outside this folder.
Overview of inbox-triage skill
What inbox-triage does
inbox-triage is an email workflow automation skill for running a structured inbox review from an existing personal email knowledge base. It searches recent mail, classifies messages using your taxonomy, identifies new senders, recommends actions, drafts replies, and writes triage learnings back to ${WORKSPACE}/Email/.
Its most important adoption detail: it is paired with inbox-setup. The inbox-triage skill expects the 7-file knowledge base that setup creates, including email-taxonomy.md, email-patterns.md, blocklist.md, tracker.md, optional decision files, and a triage-log/ directory.
Best-fit users and workflows
This skill fits users who want recurring email triage 1–3 times per day, or a reliable on-demand “check my inbox” workflow without re-explaining preferences every time. It is strongest for founders, consultants, creators, operators, and busy professionals who receive a mix of opportunities, active threads, newsletters, cold outreach, and follow-ups.
Use inbox-triage for Workflow Automation when the goal is not just summarization, but decision support: classify, recommend, draft, log, and update memory.
Key differentiators
The practical differentiator is its safety-first automation model. The skill creates drafts only and explicitly never sends email. The repository includes references/drafts_only_safety.md and scripts/draft_safety_validator.py, which checks action logs for send-shaped tool calls.
It also includes deterministic helpers: kb_reader.py validates the required knowledge base before triage, and search_window_calculator.py computes a lookback window based on cadence.
When this skill is not the right fit
Do not install inbox-triage as a standalone prompt if you have not run inbox-setup or created the expected ${WORKSPACE}/Email/ files. It is also a poor fit if you want fully autonomous sending, casual one-off email summaries, or a system that ignores user-specific taxonomy and reply style.
How to Use inbox-triage skill
inbox-triage install context
Install the skill from the GitHub skill repository with:
npx skills add alirezarezvani/claude-skills --skill inbox-triage
Then confirm that your agent can access the skill path:
productivity/email/skills/inbox-triage
Before the first run, inspect SKILL.md, then read references/kb_file_contract.md, references/drafts_only_safety.md, and references/triage_decision_framework.md. For validation behavior, review scripts/kb_reader.py, scripts/search_window_calculator.py, and scripts/draft_safety_validator.py.
Required inputs before usage
The skill needs a workspace containing ${WORKSPACE}/Email/. At minimum, the required files are email-taxonomy.md, email-patterns.md, blocklist.md, and tracker.md; triage-log/ must exist as a directory. evaluation-framework.md and rate-card.md improve decision quality for opportunities, pricing, VIPs, and pass/take signals.
If the KB is missing or malformed, the intended behavior is to halt and tell the user to run /cs:inbox-setup first. This fail-fast behavior prevents the agent from inventing triage rules.
Strong prompts that invoke the skill well
A weak prompt is: “Check my email.”
A stronger inbox-triage usage prompt is: “Run inbox-triage for my normal 2x-daily cadence. Use my existing ${WORKSPACE}/Email/ knowledge base, search only the current window, classify new messages, draft replies but do not send anything, update tracker.md and blocklist.md where appropriate, and give me the report in my preferred format.”
For on-demand runs, include the override: “Run inbox-triage for the last 24 hours because I missed the morning pass. Do not skip opportunity emails. Draft replies only.”
Practical workflow for first run
Start with a dry, supervised run. Ask the agent to validate the KB, calculate the search window, list the planned categories, and confirm no send actions will be used. Then let it process messages, create drafts, and produce the triage report.
After the report, review three things: misclassified emails, drafts that sound unlike you, and any updates written to tracker.md or blocklist.md. These corrections are high-leverage because future runs depend on the knowledge base.
inbox-triage skill FAQ
Does inbox-triage send emails?
No. The core safety rule is drafts only. The skill is designed to draft replies for user review, not transmit them. The included validator can scan a triage action log for send-like operations such as Gmail or Outlook send calls and fail the run if they appear.
How is this better than an ordinary email prompt?
A generic prompt can summarize email, but it usually lacks stable memory, sender rules, cadence windows, draft safety discipline, and a repeatable recommendation framework. The inbox-triage skill reads a defined KB contract, uses a taxonomy, applies TAKE IT / WORTH CONSIDERING / PASS / FLAG FOR REVIEW decisions, and logs outcomes.
Is inbox-triage beginner-friendly?
It is beginner-friendly only after setup. If you are comfortable running a skill installer and letting an agent access email tools and workspace files, the day-to-day invocation is simple. The initial blocker is the dependency on inbox-setup and the exact ${WORKSPACE}/Email/ file structure.
What email ecosystems can it support?
The skill is written around email-tool behavior rather than a single provider. The safety reference mentions draft-style operations for Gmail and Outlook/Microsoft Graph. Your actual success depends on whether your agent environment has email search, read, and draft-create permissions without send permissions.
How to Improve inbox-triage skill
Improve inbox-triage results with better KB files
The fastest way to improve inbox-triage is to make email-taxonomy.md and email-patterns.md more specific. Add examples of newsletters to skip, VIPs to surface, opportunities to consider, and threads that need same-day replies. Add voice rules such as “short, warm, no exclamation marks” or “include pricing only when asked.”
If opportunity triage matters, maintain evaluation-framework.md and rate-card.md. These files reduce vague “maybe” recommendations and help drafts include concrete next steps, pricing posture, or clarifying questions.
Fix common failure modes
If the skill over-drafts, tighten PASS signals and blocklist rules. If it misses important threads, expand active conversation signals and VIP entries. If drafts are too generic, add real examples of accepted and rejected replies to email-patterns.md.
If the search window is wrong, check cadence values used by search_window_calculator.py: once daily uses a broader overlap, while 2x and 3x daily use shorter windows. For unusual catch-up sessions, provide an explicit hour override.
Iterate after the first report
After each run, mark which recommendations were correct, which drafts you sent after editing, and which messages should have been skipped. Ask the agent to convert that feedback into KB updates rather than treating it as one-time chat context.
A useful follow-up prompt: “Review the last inbox-triage report. Update my taxonomy so emails like X are classified as PASS, add sender Y to VIP, and adjust my draft style to be shorter for cold outreach. Do not change send safety rules.”
Keep safety and logging enforceable
For scheduled use, keep the drafts-only rule visible in prompts and logs. Run or inspect draft_safety_validator.py if your environment supports action logs. Keep triage-log/ organized because it provides the audit trail for what was searched, drafted, skipped, and learned.