Borehole · Project · free tier
skyvern-ai/skyvern
Free reading: not for reliance
aa83f63 · 7,189 commits · 5,588 files
Read 2026-10-28

skyvern-ai/skyvern: technical due diligence, read from its repository

What Borehole read in skyvern-ai/skyvern's code and history at aa83f63: 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.

DimensionRating
Architecture & Codebase — including AI provenancestrong
Engineering Process & SDLC Maturitystrong
Product & Engineering Maturityadequate
Security & Reliability Postureadequate
Observability & Operationsadequate
Fit with the Acquirernot assessable

What the repository says

  • .gitmodules:1: 1 submodule(s) this survey does not read: integrations/n8n

Architecture & Codebase — including AI provenance

strong
01.6

skyvern/forge/sdk/workflow/models/block.py is 20,724 lines, 213× this codebase's median file

adequate

The median source file here is 97 lines. 137 files exceed 1,000 lines, the largest being 20,724. 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.

Evidence
skyvern/forge/sdk/workflow/models/block.py:1 — 20,724 lines (213× median)
skyvern/forge/sdk/workflow/service.py:1 — 19,146 lines (197× median)
skyvern/webeye/actions/handler.py:1 — 17,451 lines (179× median)
skyvern/forge/agent.py:1 — 13,125 lines (135× median)
skyvern-frontend/src/routes/workflows/copilot/WorkflowCopilotChat.tsx:1 — 11,703 lines (120× median)

Engineering Process & SDLC Maturity

strong

No finding.

Product & Engineering Maturity

adequate
04.37

API v1 and v2 have been served together for 21 months

adequate

v2 first appeared in Jan 2025, and v1 is still in the code 21 months later. Two versions mean two surfaces to secure, document and support. Ask who still calls the old one, and whether retiring it is planned.

Evidence
skyvern/forge/sdk/agent.py — v1 since Mar 2024
skyvern/forge/api_app.py — v2 since Jan 2025

Security & Reliability Posture

adequate
05.1

1 credential-shaped string in the tree sit in tests, examples, documentation or local development settings

adequate

Strings shaped like credentials appear only where fakes are usual: test fixtures, examples, documentation, placeholders and local stacks. They are almost always fakes. Ask someone to confirm that none of them works anywhere, because a real key in a development file is still a real key.

Evidence
JWT, where fakes are usual
05.6

2 Dockerfiles run the application as root

adequate

No 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.

Evidence
Dockerfile — runs as root
integrations/mcp/Dockerfile — runs as root
05.27

The build runs third-party code it does not pin

adequate

11 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.

Evidence
.github/workflows/ci.yml:45 — download piped into a shell
05.13

6 outbound calls can wait forever

adequate

6 HTTP calls use a client with no timeout (Python `requests` without `timeout=`, or Go's default client). When a dependency slows down, each one holds a worker until the connection gives up, and enough of them take the service down with the dependency.

Evidence
evaluation/core/__init__.py:46 — outbound call without a timeout
evaluation/core/__init__.py:55 — outbound call without a timeout
evaluation/core/__init__.py:62 — outbound call without a timeout
05.20

1 SQL statement built from strings

adequate

1 places in product code build SQL by formatting or joining strings rather than passing parameters. In each, what the code shows going in is a table or column name, a constant, or something the line does not show, not a request's value. Ask whether any value reaches these strings, rather than a parameter.

Evidence
skyvern/webeye/profile_download_state.py:34 — SQL built from a string
05.21

9 shell commands built from strings

adequate

9 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.

Evidence
run_streaming.py:68 — command run through a shell
skyvern/cli/database.py:34 — command run through a shell
skyvern/cli/database.py:265 — command run through a shell
05.22

3 deserialisers that can run code

adequate

3 places read data with a deserialiser that executes what it reads (pickle, `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.

Evidence
skyvern/forge/sdk/utils/capped_subprocess_child.py:46 — code-executing deserialiser
skyvern/forge/sdk/utils/capped_subprocess_child.py:48 — code-executing deserialiser
skyvern/utils/yaml_loader.py:56 — code-executing deserialiser
05.26

5 locked dependencies have published vulnerabilities, 2 high or critical in what ships

adequate

5 of 1,121 locked dependencies match 5 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.

Evidence
skyvern-frontend/package-lock.json — body-parser 1.20.6: GHSA-27r3-jxmh-p828 (high), fixed in 1.20.9
skyvern-frontend/package-lock.json — zod 3.25.76: GHSA-mrhx-c7xc-m3w5 (high), fixed in 4.6.6
uv.lock — litellm 1.89.7: GHSA-g5ff-637f-6q2m (high), fixed in 1.95.0; an optional extra
05.48

6 places insert HTML from a variable without sanitising it

adequate

6 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.

Evidence
site_fixtures/code_host/index.html:246 — HTML from a variable
site_fixtures/code_host/index.html:276 — HTML from a variable
site_fixtures/code_host/index.html:315 — HTML from a variable
05.65

2 advisories against this project in the last 24 months, 2 high or critical

adequate

The newest was published on 2026-08-29. A project that publishes advisories is telling its users about its fixes, and the count and pace say how much security work it carries. None of them affects the surveyed version.

Evidence
2026-08-29: CVE-2026-82447 (high): Skyvern before 1.0.45 contains a sandbox escape vulnerability in TextPromptBlock that renders prompts twice, first through a sandboxed Jinja environment and the
2025-06-07: GHSA-h92g-3xc3-ww2r (high): Skyvern has a Jinja runtime leak

Observability & Operations

adequate
06.9

No error reporter is wired in

adequate

Nothing 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.

Evidence
no error reporter initialised
06.14

Readiness and liveness probes are identical in 1 of 1 workloads

adequate

Both probes call the same endpoint. When a dependency slows down, the liveness probe fails with the readiness probe, so the platform restarts pods that only needed taking out of rotation, and a partial outage becomes a full one. Liveness should ask whether the process is alive; readiness whether it can serve.

Evidence
k8s/charts/skyvern-ui/templates/deployment.yaml — same probe endpoint twice
06.33

3 of 4 deployments run a single replica

adequate

3 deployments set `replicas: 1`, and nothing (an autoscaler, a disruption budget) adds more. Every deploy and every node drain is downtime, and the availability the buyer inherits has a ceiling.

Fit with the Acquirer

not assessable

Not rated: structurally not assessable from a repository.

Read more, or read your own

The full report on skyvern-ai/skyvern: 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.