Read 2026-10-23
nousresearch/hermes-agent: technical due diligence, read from its repository
What Borehole read in nousresearch/hermes-agent's code and history at 290080e: its licence, security advisories, releases, review practice, tests, dependencies and operations. Every finding cites the file and line it rests on. The full report also covers the team and asks the questions the code cannot answer.
| Dimension | Rating |
|---|---|
| Architecture & Codebase — including AI provenance | adequate |
| Engineering Process & SDLC Maturity | strong |
| Product & Engineering Maturity | adequate |
| Security & Reliability Posture | adequate |
| Observability & Operations | adequate |
| Fit with the Acquirer | not assessable |
Architecture & Codebase — including AI provenance
adequateapps/desktop/electron/main.ts is 18,333 lines, 235× this codebase's median file
adequateThe median source file here is 78 lines. 284 files exceed 1,000 lines, the largest being 18,333. Size is not a defect on its own — some problems genuinely live in one place — but these are where merge conflicts, review fatigue and single-owner knowledge concentrate, and they are the first thing that slows an inheriting engineer down.
apps/desktop/electron/main.ts:1 — 18,333 lines (235× median) |
agent/auxiliary_client.py:1 — 8,393 lines (107× median) |
plugins/platforms/discord/adapter.py:1 — 7,415 lines (95× median) |
plugins/platforms/telegram/adapter.py:1 — 7,382 lines (94× median) |
plugins/platforms/slack/adapter.py:1 — 6,956 lines (89× median) |
1,110 errors caught and discarded in product code
adequate1,110 places catch an error and do nothing with it, across 450 files (0.6 per thousand lines). Each is a failure the system will not report. They surface later as problems nobody can explain, and they are cheap to fix now.
acp_adapter/auth.py:25 — error caught and discarded |
acp_adapter/tools.py:149 — error caught and discarded |
acp_adapter/tools.py:688 — error caught and discarded |
1 model weight file committed directly to git
adequate1 model weight files are stored in git itself, not in LFS, DVC or a model hub. Every clone carries them, the history keeps every version for good, and nothing records which run or data produced them.
contributors/emails/ada.a0uew@56.pt — weights in git |
Engineering Process & SDLC Maturity
strongNo finding.
Product & Engineering Maturity
adequate82 direct dependencies in a single manifest
adequateapps/desktop/package.json declares 82 direct dependencies. Each one is a maintenance obligation, a supply-chain surface and a potential licence question that transfers with the asset. The count says nothing about whether any given dependency was a good choice; it says how many such choices a buyer inherits.
apps/desktop/package.json — 82 declared |
pyproject.toml — 46 declared |
web/package.json — 23 declared |
apps/bootstrap-installer/package.json — 20 declared |
Security & Reliability Posture
adequate2 credential-shaped strings committed to the tree
adequateMatches for Slack token appear in tracked files outside test and fixture paths. They stay in the history even after they leave the tree, so what matters is whether each was rotated. The code cannot tell whether any of them is live, and that decides whether this matters: none sits in production configuration, and none is in a format its provider keeps for live keys. Is any of these live? Show that each was rotated after it was committed, and when.
| 2 cited: Slack token pattern matched |
1 Dockerfile runs the application as root
adequateNo USER instruction switches away from root, so the process runs with full privileges inside the container. It is a one-line fix and an ordinary expectation, which is why its absence is worth asking about: it usually says more about whether anyone has reviewed the deployment than about this container specifically.
Dockerfile — runs as root |
The build runs third-party code it does not pin
adequate4 steps pipe a download straight into a shell. Tags have been moved to malicious commits in real supply-chain attacks, and a piped script is whatever the server returns that minute. Pinning to a commit is the fix, and it is small.
.github/workflows/bootstrap-installer.yml:42 — download piped into a shell |
148 SQL statements built from strings
adequate148 places in product code build SQL by formatting or joining strings rather than passing parameters, and in 65 of them a value, not a table or column name, goes into the string. Where any of those values comes from a request, it is an injection point. Each site needs a look; the fix, parameters, is small.
agent/transcript_repair.py:74 — SQL built from a string |
cron/executions.py:596 — SQL built from a string |
cron/incidents.py:272 — SQL built from a string |
19 shell commands built from strings
adequate19 places run a command through a shell (`shell=True`, `os.system`, `exec` with a template) rather than passing arguments. Where any part comes from outside, it is command injection. Ask where each one's input comes from.
agent/command_token_source.py:55 — command run through a shell |
apps/desktop/electron/main.ts:8840 — command run through a shell |
cli.py:1251 — command run through a shell |
5 deserialisers that can run code
adequate5 places read data with a deserialiser that executes what it reads (`yaml.load` without a safe loader). That is safe on data the system wrote itself and remote code execution on anything else. Ask where each one's input comes from.
plugins/security-guidance/patterns.py:44 — code-executing deserialiser |
plugins/security-guidance/patterns.py:44 — code-executing deserialiser |
plugins/security-guidance/patterns.py:146 — code-executing deserialiser |
3 locked dependencies have published vulnerabilities, 2 high or critical in what ships
adequate3 of 2,209 locked dependencies match 4 published advisories in OSV, 3 of them rated high or critical. A lockfile says which version is installed, not whether the flawed code is reachable from this product, so each needs a look. Most are fixed by an upgrade, and the advisory names the version.
uv.lock — brotlicffi 1.2.0.2: GHSA-9v8h-qqgm-4m35 (critical), fixed in 1.2.1.0; an optional extra |
package-lock.json — zod 4.4.3: GHSA-mrhx-c7xc-m3w5 (high), fixed in 4.6.6 |
plugins/platforms/photon/sidecar/package-lock.json — zod 4.5.4: GHSA-mrhx-c7xc-m3w5 (high), fixed in 4.6.6 |
2 workload settings can give a container the host, behind a chart value or in a stack built from source
adequate2 places run a container privileged or in the host's namespaces only when a Helm value switches them on, or in a compose file that builds the product from this repository, which is how it is developed rather than how it ships. Neither is on by default in what ships. Ask whether production sets the value or runs that stack.
docker-compose.yml:35 — host access behind a setting or in a local stack |
docker-compose.yml:67 — host access behind a setting or in a local stack |
10 places insert HTML from a variable without sanitising it
adequate10 places set HTML from a value that is neither a literal nor passed through a sanitiser. Where the value can come from a user, that is stored cross-site scripting, and it is the first thing a buyer's security review looks for in a frontend.
apps/desktop/src/app/chat/right-rail/preview-artifact.tsx:91 — HTML from a variable |
apps/desktop/src/components/assistant-ui/embeds/svg-embed.tsx:29 — HTML from a variable |
apps/desktop/src/components/chat/shiki-block.tsx:175 — HTML from a variable |
2 vendored components with no licence
adequateThird-party code is committed into this repository without any statement of the terms it arrived under. Unknown terms are harder to deal with than inconvenient ones: an awkward licence can be complied with, an absent one cannot. Each of these needs its origin established before the code changes hands.
native/fts5_cjk/vendor/sqlite3.h — 1 tracked files, no licence file |
native/fts5_cjk/vendor/sqlite3ext.h — 1 tracked files, no licence file |
15 advisories against this project in the last 24 months, 4 high or critical
adequateThe newest was published on 2026-10-21. A project that publishes advisories is telling its users about its fixes, and the count and pace say how much security work it carries. Whether 3 of them are fixed in the code you ship could not be told from this history: its version is only known to be at least 2026.9.24. Answer lives in: the release you run, set against each fixed version. The record is incomplete: NVD: 6 records name hermes-agent with nothing tying them to this repository (CVE-2026-10548, CVE-2026-108593, CVE-2026-18774, …).
| 2026-10-21: CVE-2026-114463 (unrated): The webhook platform gateway in Nous Research Hermes Agent versions builds the session chat_id as webhook:{route_name}:{delivery_id} and re-derives the authoriz |
| 2026-09-03: CVE-2026-71963 (high): Hermes Agent 0.18.2 through 0.21.0, fixed in commit f6234d0, contains a remote code execution vulnerability that allows attackers to execute arbitrary OS comman |
| 2026-08-28: CVE-2026-82020 (high): Hermes Agent 0.16.0 prior to 0.17.0 contains an improper path restriction vulnerability that allows attackers who can influence ingested message content to over |
| 2026-08-28: CVE-2026-82021 (critical): Hermes Agent 0.18.2 prior to 0.19.0 contains a supply chain vulnerability in its bundled MCP catalog that allows a remote attacker to execute arbitrary code by |
| 2026-08-06: CVE-2026-18976 (low): A vulnerability was determined in NousResearch hermes-agent up to 0.16.0. This impacts the function get_tool_definitions of the file agent/agent_init.py of the |
Observability & Operations
adequateNo error reporter is wired in
adequateNothing in the service initialises an error reporter (Sentry, Rollbar, Bugsnag, Honeybadger or an APM agent). An exception in production reaches a person only if someone is reading the logs, which usually means a customer reports it first.
| no error reporter initialised |
Sends usage data to telemetry.nousresearch.com by default
adequatehermes_cli/config_defaults.py sends to telemetry.nousresearch.com, and agent/context_compressor.py leaves reporting on as shipped. Every installation reports to the vendor unless its operator turns this off. For a buyer running it inside their own product, that is an outbound data flow to a third party: ask what is sent, how often, and whether the buyer's customers were told.
hermes_cli/config_defaults.py:2335 — reports to telemetry.nousresearch.com |
agent/context_compressor.py:3792 — the reporting setting is on by default |
2 of 2 deployed images carry no version, or `latest`
adequate2 images in the deployment configuration are named without a version or as `latest`. A redeploy can change the software without a commit, a rollback has nothing to roll back to, and nobody can say from the repository what is running.
docker-compose.windows.yml:25 — image without a version |
docker-compose.yml:64 — image without a version |
Fit with the Acquirer
not assessableNot rated: structurally not assessable from a repository.
Read more, or read your own
The full report on nousresearch/hermes-agent: every dimension, the questions for the room, and the audit log.
Survey a repository: free on public repositories. Private code is read on your own machine by a collector whose source you can read first.