Install
$ agentstack add skill-shiftedx-hermes-agent-blueprint-hermes-agent-blueprint ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo 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.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
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 →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:
- Model and provider — the owner runs
hermes setupthemselves. Nothing here depends on it. - Persona —
templates/SOUL.mdships 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.
- Autonomy level —
reference/02is 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.
- 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.
- Topic taxonomy —
scripts/lucid.config.json'stopicsmap 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.
- 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-sizebefore 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.
- Author: shiftedx
- Source: shiftedx/hermes-agent-blueprint
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.