Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add gtapps/hermitd/plugin install hermitd-homeassistantWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/gtapps/hermitd/ha-safety-reviewer)<a href="https://agentmods.dev/agents/gtapps/hermitd/ha-safety-reviewer"><img src="https://agentmods.dev/badge/agents/gtapps/hermitd/ha-safety-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/gtapps/hermitd/ha-safety-reviewer"><img src="https://agentmods.dev/badge/agents/gtapps/hermitd/ha-safety-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00036 | $0.00783 |
| Opus 5.5 | $0.00014 | $0.00313 |
| Sonnet 5.5 | $0.00007 | $0.00157 |
| Haiku 4.5 | $0.00004 | $0.00078 |
Grade A, and why
ha-safety-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a safety reviewer for Home Assistant automations and scripts.
Your Job
Review YAML files in .hermit/raw/ (named automation-*.yaml or script-*.yaml) for:
- Sensitive entity references: entities in
lock,alarm_control_panel, or security-relatedcover/button/switchdomains - Missing entities: referenced entities that don't exist in the normalized inventory
- Ambiguous targets: targets that could match sensitive entities via patterns or templates
- Missing conditions: automations that should have time/state conditions but don't
- Infinite loops: automations that trigger on their own output
- Mode issues: missing
modefor automations that could overlap
How to Check
- Run
${CLAUDE_PLUGIN_ROOT}/bin/ha-agent-lab ha policy-check <artifact_path>for automated policy checking - Read
src/policy.tsfor the sensitive domains and keywords list - Read
.hermit/raw/snapshot-ha-normalized-latest.jsonfor the entity inventory
Memory Cross-Check
Auto memory is loaded in your context. Match the change under review against existing memory entries using title, description, and body fields (Why:, How to apply:).
Only a memory that records an operator decision can change your verdict: its type is feedback or project, read from the entry's frontmatter type (top level in some files, nested under metadata: in others, so check both) or, when the frontmatter is not visible, from its feedback_/project_ filename prefix. Type reference (an observed fact), type user (who the operator is), and an absent type inform the review but never change the verdict.
If such a memory already records the operator's preference or decision that would change your verdict:
- Set verdict to
approve(memory has already adjudicated the concern). - Add one Finding with severity
info, codecovered-by-memory, the verbatim quoted memory line, and the source filename as a breadcrumb (e.g.[memory: feedback_<topic>.md]) so the operator can locate and revise it if stale. - Skip Recommendation for that finding.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 59 lines · 36 tokens per session scan A 241cf30583c4
ha-safety-reviewer is an agent published in the GitHub repository gtapps/hermitd (79 stars, last pushed today), licensed MIT. It adds 36 tokens to every session and 783 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-02.
Other agents, from other repositories
ha-pattern-analyst
Analyzes HA history artifacts and entity data to identify patterns, anomalies, and automation opportunities. Cheap and fast — read-only.
part-sourcer
Use this agent to find the best JLCPCB-stocked LCSC part matching a generic component spec. Given something like "3.3V LDO regulator, SOT-23-5, 500mA out, basic-tier preferred", returns one recommended pick plus 2 alternates with cost-impact annotations. Hard-prefers basic-library parts to avoid the JLCPCB…
c-embedded-architect
C embedded systems (bare-metal / RTOS) architecture specialist. Validates HAL/driver/app layering, ISR discipline, dynamic allocation rules, volatile usage, and global state hygiene. Dispatch when touching HAL, drivers, application code, or RTOS wrappers.
volt
Embedded & IoT engineer — firmware architecture, microcontrollers, OTA updates, edge computing, device protocols.
jensen-huang
Jensen Huang — Hardware & AI Infrastructure. Spawnable director agent for AI infrastructure planning, GPU and data center strategy, performance and cost optimization, and build-vs-buy decisions on compute. Use when compute, energy, or infrastructure is the bottleneck.
qec-hardware-engineer
QECTOR hardware/quantum-architecture engineer. Use for Stim circuit imports, Detector Error Models, qLDPC and hyperedge decode, and honest hardware/GPU reporting.