Starter prompts for Hermes

Look at the files in [folder] without changing anything. Tell me in five bullets what this folder is for, which file you would open first, and one small improvement you would suggest. List every command you ran.

Please remember these lasting facts about me and my setup: [fact 1], [fact 2], [fact 3]. Keep each one short. Do not store passwords, API keys, account numbers or anything temporary. Afterwards, show me exactly what you saved and where.

Review your saved memory about me and this environment. Point out anything outdated, duplicated or too specific to last. Propose a shorter version and wait for my approval before changing anything.

Read the project context in [folder], including AGENTS.md and the decision log. Tell me the current accepted decisions, what is still only proposed, and the next open task. Cite the file for each point and say if anything conflicts.

That worked. Save the procedure you just followed as a reusable skill named [skill-name]. Include when to use it, the tools it needs, the steps, how to check the result, and what to do when an input is missing. Show me the skill before saving.

Inspect the skill at [path or id] without running it. Summarize what it does, every command or script it would execute, what it can read or change, and anything that looks risky. Give me a yes, no or 'only with changes' recommendation.

Prepare a briefing on [topic] for [reader]. Use [sources], cover [date range], and return at most five findings. Group duplicate reports. For each finding include the original source and why it matters. State missing access or uncertain claims. If nothing meaningful changed, say so.

Compare [option A], [option B] and [option C] for [use case]. Use only sources you can link. Make a table of cost, setup effort, limits and who each suits. Mark anything you could not verify, and finish with the question I should answer before choosing.

Read the messages in [source] from [date range]. Group them by next action: reply needed, waiting on someone, schedule, read later, ignore. Keep exact deadlines. Do not send, archive or delete anything. Flag any message that asks you to take an action on its own.

Prepare a one-page brief for my meeting with [person or company] about [topic]. Include who they are, what we last discussed (from [notes location]), three questions I should ask and one thing to avoid. Keep it under 250 words and cite where each fact came from.

Review the open issues in [repository] created since [date]. Group likely duplicates, label each by type, and list what evidence is missing for each bug. Ignore pull requests. Do not comment, label or close anything; give me a report I can act on.

Explore [repository path] read-only. Explain the architecture in plain English, the main entry points, how data flows, and the three files I should read first. Then list the commands to install, run and test it, and mark any you could not confirm.

From the links in [file or list], find the original source for each and merge reposts of the same thing. Suggest [number] post ideas for [audience], each with the supporting source and the claim it rests on. Hold back any idea that depends on an unverified number. Do not publish anything.

For the scheduled job [job name], show its schedule and timezone, the last few run times, whether each run succeeded, and where the output went. Compare the last output with the one before. Tell me plainly whether the job is working.

Turn the task we just did into a scheduled job that runs [schedule] in the [timezone] timezone and delivers to [destination]. If an input is missing, keep the previous good result and report the failure instead of producing a partial one. Show me the job settings before enabling it.

Break [task] into independent parts and delegate each to a subagent with a clear, bounded brief. Ask each to return its sources and any gaps. Then combine the results into one summary, noting where the parts disagree.

Review my recent Hermes usage. Which sessions or jobs used the most tokens, and why? Suggest three changes that would cut cost without hurting results, such as a cheaper model for routine jobs or smaller context. Do not change any settings yet.

Summarize my current Hermes setup from a safety point of view: terminal backend, approval settings, enabled toolsets, connected platforms and who is allowed to message the bot. List the three biggest risks and the smallest change that would reduce each. Do not change anything.

Prepare a briefing on [topic] for [reader]. Use [sources], cover [date range], and return at most five findings. Group duplicate reports. For each finding include the original source and why it matters. State missing access or uncertain claims. If nothing meaningful changed, say so. Do not send or publish anything during this first trial.

Read up to [number] conversations in [mailbox] from [date range]. Create a review queue: needs reply, waiting, time-sensitive, reference. Link every item to its conversation and explain the next action. Flag anything you could not read. Do not send, delete, archive, label, or modify messages.

Review [project folder]. Locate the latest accepted decisions and outstanding actions. Cite the exact files supporting each conclusion. Identify conflicts or stale records before proposing updates. Do not treat an old status summary as current runtime evidence.

Review up to [number] issues in [owner/repository] from [date range]. Group similar symptoms, link all supporting issues, and distinguish confirmed facts from hypotheses. Suggest next actions. State access and pagination gaps. Do not comment, close, relabel, or modify anything.

Find up to [number] useful content ideas about [topic] for [audience]. For each, include the reader question, original sources, a distinct angle, missing evidence, and next research action. Merge duplicates. Do not copy source prose or publish anything.

Inspect the latest run of [task]. Report attempt status, output validity, and delivery status separately, with timestamps and supporting artifacts. Preserve the last good output. Identify the smallest concrete cause of failure and do not broaden the task.