AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Hermes Agent Blueprint

skill-shiftedx-hermes-agent-blueprint-hermes-agent-blueprint · by shiftedx

Use when setting up, hardening, or debugging a Hermes Agent deployment — autonomy and approval posture, skill-set trimming, the holographic fact store, an Obsidian vault brain, and a nightly self-review ("Lucid Dreaming") cron cycle. Also use when a Hermes cron job, memory recall, or scheduled review has silently stopped working.

No reviews yet
0 installs
18 views
0.0% view→install

Install

$ agentstack add skill-shiftedx-hermes-agent-blueprint-hermes-agent-blueprint

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-shiftedx-hermes-agent-blueprint-hermes-agent-blueprint)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
1mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Hermes Agent Blueprint? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Hermes Agent blueprint

Stands up a Hermes Agent profile with durable memory, a browsable Obsidian brain, and an unattended nightly review cycle — or diagnoses one that has quietly stopped working.

Everything here is version-sensitive. Run reference/00-verify-your-build.md first. It is a short list of commands that check every assumption this blueprint makes against the build in front of you. Do not paste config from the later files until it passes.

Before anything else

Read reference/00-verify-your-build.md and run it. It takes about two minutes and it is the difference between this blueprint being accurate and being confidently wrong about a newer Hermes.

Decisions the owner makes, not you

Interview them before writing files. Wrong guesses here are expensive to undo:

  1. Model and provider — the owner runs hermes setup themselves. Nothing here depends on it.
  2. Personatemplates/SOUL.md ships a teaching-assistant posture: does the real work at

full capability, but narrates the reasoning so the human learns from watching. Swap it if that is not what they want. Keep it under ~1KB; it is injected every turn.

  1. Autonomy levelreference/02 is a deliberately high-trust posture: local shell,

real filesystem, no per-command prompts for ordinary work. The ⚑ guardrails are what make that survivable. Tighten it for a shared or minor's machine; do not tighten it silently.

  1. What fires the schedule — a gateway service, or an OS timer calling hermes cron tick.

See §8a. Getting this wrong is the single most common silent failure.

  1. Topic taxonomyscripts/lucid.config.json's topics map drives both trend detection

and the vault graph. The shipped defaults are generic on purpose. Ten minutes matching them to the owner's real work is the difference between a useful memory graph and a grey blob.

  1. Where the workspace lives — anywhere outside iCloud/Dropbox/OneDrive. Sync daemons race

the vault's git repo and the SQLite files.

Files, and the § numbers they contain

Cross-references throughout use §N. Resolve them here:

| Read | Sections | Covers | |---|---|---| | reference/00-verify-your-build.md | — | version preflight; run this first | | reference/01-foundation.md | §0–§3 | variables, install, profile layout, the four identity files | | reference/02-autonomy-and-capability.md | §4–§5 | approvals, guardrails, skill trimming | | reference/03-memory-and-vault.md | §6–§7 | fact store schema, retrieval, the Obsidian brain | | reference/04-lucid-dreaming.md | §8 | the nightly cycle, cron prompts, ⚑ §8a scheduling | | reference/05-operations.md | §9 | staying alive, safe config edits, log rotation | | reference/06-bring-up-and-owner-handoff.md | §11–§13 | the checklist; what to tell the human | | reference/example-deployment.md | §10 | one real deployment's numbers and corrections |

templates/ holds SOUL.md, AGENTS.md, and the two cron prompts — all with ` blanks. scripts/ is the nightly cycle, ready to install; see scripts/README.md`.

Working rules

  • Verify, don't assert. Every checkbox in §11 needs a command and its output. Most of the

corrections in example-deployment.md exist because earlier notes asserted state instead of re-reading it — including one where "we patched X locally" was wrong because upstream had adopted the fix months earlier.

  • Read §8a before building any part of §8. Cron jobs are fired by the gateway's background

ticker. Create the whole nightly cycle without one and every job is created successfully, appears in the Cron pane, and never runs. There is no error message.

  • Prove the schedule with a real job. "Task registered" and "jobs run" are different claims.

Create a throwaway job due in two minutes, confirm output lands in $HERMES_ROOT/cron/output/ and a row appears in cron/executions.db, then delete it.

  • Measure the prompt budget. hermes prompt-size before and after skill trimming; record

both numbers. Trimming is the highest-leverage tuning knob here.

  • Test destructive SQL on a copy of the database. Never the live one.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.