Weekly Review Planning: Run a Bounded Weekly Reset with Hermes Agent

The Weekly Review Planning skill turns a recurring, often vague ritual into a bounded, repeatable process. You hand it a request like "Run my weekly review" or "What did I commit to and what is slipping?" and it walks through a fixed sequence: confirm your systems, gather evidence from calendar and notes, clear inboxes, reconcile projects, chase waiting items, build a plan that fits your actual capacity, and apply only the updates you approve. This is a concrete task, not a generic methodology. The weekly-review Automation Blueprint schedules it as a cron job, so it can fire on a timer without you typing anything.

Reach for this when you want a disciplined, repeatable review that ends with a clear plan and a list of proposed changes. It suits anyone who juggles multiple projects, a calendar, and a task system, and who wants to catch stalled work before it becomes a problem.

What it does

The skill runs a seven-step procedure that mirrors a classic weekly review but with guardrails. It starts by confirming your timezone, the review period, the planning horizon, and which stores are authoritative for tasks, projects, calendars, and inboxes. It then pulls evidence from those systems, proposes a plan, and only mutates anything after you approve the scope. The output is a structured summary: wins, overdue items, waiting items, stalled projects, next week's outcomes, proposed updates, and coverage gaps.

The key design choice is that the skill defaults to recommendations and drafts, not direct changes. It will not silently delete or reschedule anything. Every proposed update is listed for approval, and once approved, the skill writes the changes and reads them back to verify they stuck.

Before you start

This skill is bundled with Hermes Agent, so it is installed by default. It lives at skills/productivity/weekly-review-planning and is version 0.1.0. It runs on linux, macos, and windows. The author is Ben Barclay (benbarclay) for Hermes Agent, and it is released under the MIT license.

Before the first run, make sure the connectors it relies on are available. The skill references google-workspace for calendar evidence, obsidian and notion for notes, and email-inbox-triage for flagged email. Those are separate skills; this one orchestrates them rather than replacing them. You will also need to confirm which task or project store is authoritative, because the skill needs a declared winner when sources conflict.

Procedure

The skill follows a fixed order. Each step has a clear completion condition.

1. Set systems and window

Confirm timezone, review period, planning horizon, authoritative task/project store, calendars, inboxes, and allowed writes. Default to recommendations/drafts, not mutations. Done when source-of-truth conflicts have a declared winner.

This step is about agreeing on the ground rules before touching any data. If you have multiple task lists, decide which one is the source of truth. The skill will not guess.

2. Review calendar evidence

Load google-workspace or the relevant calendar connector. Inspect the completed week for meetings and commitments, then the next 1-2 weeks for deadlines, travel, preparation, and capacity. Capture follow-ups implied by past events and conflicts ahead. Done when both retrospective and horizon are covered.

You are looking for two things: what actually happened last week that needs a follow-up, and what is coming up that needs preparation. The skill will not move on until both the past week and the next two weeks have been examined.

3. Clear capture inboxes

Review the task inbox, notes (obsidian, notion), flagged email (email-inbox-triage owns thread-level triage), and other declared capture points. Convert each item to next action, project, waiting, scheduled, someday, reference, archive, or delete proposal. Do not mutate until scope is approved. Done when remaining unprocessed items are counted and stated.

This is triage, not cleanup. Every item gets a proposed disposition, but nothing is actually moved or deleted until you approve. The skill will tell you how many items remain unprocessed.

4. Reconcile active projects

For each project identify desired outcome, next action, owner, deadline, blocker, last meaningful activity, and source link. Flag projects with no next action, missed dates, duplicate records, or contradictory status. Done when every active project is actionable or explicitly paused.

This step surfaces the projects that look alive but are actually stalled. A project with no next action is a red flag. The skill will mark it as such.

5. Review waiting and commitments

Find promises made by the user and items owed by others. Propose follow-ups with dates and channels. Do not infer that silence means completion. Done when each waiting item has an owner and next review/follow-up date.

A common failure is assuming that no news is good news. This step forces a follow-up date and channel for every waiting item, so nothing drifts.

6. Build a capacity-aware plan

Estimate fixed calendar load and select a small set of weekly outcomes plus near-term next actions. Rank by consequence, deadline, dependency, and effort; do not fill every free hour. Done when the plan fits actual capacity and names deferred work.

The plan is deliberately small. It ranks by consequence, deadline, dependency, and effort, and it leaves slack. It also names what is being deferred, so you know what you are not doing.

7. Apply approved updates

Update tasks/projects, create calendar holds, archive processed items, and draft follow-ups only as approved. Read every changed record back from the provider. Done when verified writes match the review summary.

Only after you approve does the skill write anything. It then reads back each changed record to confirm the write actually happened and matches what was intended.

Output Shape

The skill produces a structured summary with these sections:

1. Wins and completed commitments
2. Overdue or at risk
3. Waiting/follow-ups
4. Stalled or ambiguous projects
5. Next week's outcomes and calendar constraints
6. Proposed updates awaiting approval
7. Coverage gaps

This shape makes it easy to scan the review and see what needs attention.

When not to use it

The source explicitly says not to use this for daily briefs or single-inbox triage. For a daily brief, see the google-workspace daily-brief reference. For triaging a single inbox, use email-inbox-triage. This skill is for the weekly cadence, not the daily one.

Limits and gotchas

The source lists several pitfalls to avoid:

• Planning from tasks without calendar capacity.
• Carrying every unfinished item forward as high priority.
• Marking projects active with no next action.
• Silently deleting or rescheduling personal commitments.
• Treating silence from others as completion.

The skill is designed to prevent these, but you still need to review its proposals. It will not automatically know that a project is truly done or that a commitment is no longer relevant.

Verification

The skill checks its own work against these criteria:

• Both the completed week and the planning horizon were covered, or gaps are stated.
• Every stalled/waiting flag traces to a specific record, event, or thread.
• No task, event, or note was mutated without approval; approved writes were read back.
• The plan names what was deferred, not just what was chosen.

If any of these fail, the review is incomplete.

What pairs with this

The source lists related skills: obsidian, notion, airtable, google-workspace, and email-inbox-triage. Use those to feed data into the review and to act on its output. The weekly review is the orchestration layer; the others are the systems it reads from and writes to.