1. LandOffer
  2. Blog
  3. All articles
  4. Before You Run a Coding Interview Repository: Verify the Request and Inspect Execution Boundaries

Job search

Before You Run a Coding Interview Repository: Verify the Request and Inspect Execution Boundaries

A completed static configuration audit and an unsent changed-request clarification separate read-only review, execution and verified recruiting authority.

9 min readLandOffer team

Before running a coding-interview repository, verify who requested it, what exact project and commands they expect, and which permissions those commands would receive. Review the instructions and execution-related configuration before installing dependencies or trusting the workspace. A familiar logo, public repository or completed download does not verify the recruiting request.

This guide checks current VS Code, npm, GitHub and FTC documentation on October 21, 2026. Its two original cases are fictional; the only executed exercise is a local read-only analysis of synthetic configuration text. No foreign repository, archive, installation hook or project command was run. LandOffer publishes this guide and offers job-search tools; you can browse recent technical roles while keeping interview preparation tied to verified employer instructions.

A developer pauses to inspect an interview request and configuration before connecting an execution path

Verify the request before treating the repository as an assignment

The FTC's December 2023 job-scam guidance, still publicly available when checked, warns that realistic interviews and official-looking materials can accompany scams. It recommends contacting the company through a website or phone number independently known to be legitimate, rather than relying on contact details supplied by the purported recruiter. Requests for money or sensitive personal information are additional warning signs.

Apply that verification principle to a repository handoff. Compare the sender, role, interview round and repository owner with the recruiting conversation you already trust. If a new sender replaces the agreed exercise with an urgent download, ask your verified recruiting contact to confirm the change. A repository name that resembles the company is one piece of context, not proof of ownership or authorization.

Record the specific assignment: repository URL or delivered filename, intended branch or version if supplied, expected runtime, permitted tools and submission route. Do not infer an undocumented requirement from a README that asks for broader access than the invitation describes. The content under review cannot independently authorize itself to use your personal keys, accounts or administrator privileges.

This check can reveal an ordinary process mismatch as well as a malicious request. A recruiter may have sent the wrong exercise or an outdated environment guide. Describe the discrepancy without asserting fraud that you have not established. The useful outcome is a confirmed assignment or a paused execution decision, not an accusation based only on unfamiliar tooling.

Identify what setup would execute

Start with the files that explain how the project expects to operate. Read the README's setup and test sections, then trace the named scripts to their definitions. Look at dependency manifests and lockfiles, editor tasks and settings, development-container configuration, and agent or tool-server configuration when present. The question is what an action would start, not whether a filename looks conventional.

For each proposed action, identify the executable, arguments, working directory and inputs. Ask whether it installs packages, invokes another script, accesses a network service, requests a credential or starts something automatically. A short setup command can delegate to a longer chain. Do not describe that chain as reviewed when you have only read its first line.

Configuration review is incomplete if referenced files are missing or dependencies introduce behavior you have not inspected. Name the missing script or dependency instead of treating an absent description as proof of no execution. A lockfile can help identify declared dependencies; it does not establish that every package or downloaded artifact is trustworthy.

This article does not provide a universal scanner or a safe command to run every interview project. Its static example shows how to turn visible configuration into specific questions before execution. You may reasonably need the employer to supply an approved environment or corrected instructions before proceeding.

Restricted Mode is a review aid with limits

The current VS Code Workspace Trust documentation describes Restricted Mode as disabling or limiting AI agents, terminals, tasks, debugging, workspace settings and extensions while unfamiliar content is reviewed. It says trust is shared between the VS Code window and the Agents window. Trusting one surface should therefore not be treated as a separate harmless choice about appearance.

The same documentation cautions that a malicious extension can ignore Restricted Mode. Folder trust can also be inherited from a trusted parent. Restricted Mode is not an operating-system sandbox, a malware verdict or a guarantee that every installed component respects its boundary. Check the visible trust state rather than assuming a newly opened folder is restricted merely because its contents are unfamiliar.

Keep review and execution separate in your own description. “I read the setup instructions” does not mean “I approved all commands.” If opening a terminal or task asks you to trust the workspace, treat that as a decision to evaluate, not an obstacle to dismiss quickly so the interview can start.

Do not disable protections or install a new extension simply because the unverified project instructs you to do so. When the employer requires a specific approved setup, resolve the access and environment questions through its verified channel. The goal is to complete the legitimate exercise within an understood boundary.

Worked static audit: a simple README hides several execution entry points

Fictional candidate Amara receives a hypothetical project described as a small JSON transformation exercise. The original synthetic README says to install dependencies and run tests, and describes setup as straightforward. The teaching fixture contains only inert text copies of configuration; no referenced script implementation is provided or executed.

Its package manifest names postinstall as node scripts/prepare.cjs and test as node tests/check.cjs. The editor task points to npm run setup and declares a folder-opening trigger. The development-container configuration also names npm run setup as a post-creation command. A tool-server configuration names node tools/context.cjs. These strings are evidence of proposed execution, not findings that the named scripts are harmful.

Inspected location Completed static finding Amara's decision
README and package manifest The install path has a declared lifecycle script Read its implementation and applicable package-manager policy before installation
Package test entry A test command explicitly targets a script Do not classify tests as nonexecuting inspection
Editor task A task requests execution on folder opening Review the trigger and trust requirement before enabling it
Development-container file Creation requests a setup command Do not treat creating the environment as a passive preview
Tool-server configuration A local executable is named Clarify whether the interview requires that tool and what it can access

An original Python analyzer loaded these four synthetic JSON documents as data in the completed local check and extracted those five entries. It checked the expected script names and trigger without invoking any of them. The saved results distinguish an install lifecycle entry, an explicit test entry, a folder task, a container setup command and a tool-server command. No package manager, container, editor task or tool server was started.

The result is execution review incomplete. Amara has identified the next files and permissions to inspect, but she has not proved safety, maliciousness or compatibility. The analyzer is deliberately narrow: it does not interpret every configuration format, recursively inspect dependencies or detect obfuscated behavior. It does not establish whether a real repository can be run safely.

This distinction makes the record useful. Instead of saying “the repo looks suspicious,” Amara can explain which action leads to which unreviewed script. If the intended assignment is only to read and discuss a function, she can also ask whether any of those setup paths are necessary at all.

Understand what an install flag does and does not change

The current npm ci documentation, displaying CLI 12.2.0 when checked, describes clean project installation from a lockfile. It also documents script controls. With ignore-scripts enabled, npm skips scripts specified in package manifests, but explicitly requested commands such as npm test, npm start or npm run still run their target script, without its pre- and post-scripts.

The same current page documents install-script policy through allowScripts and related configuration. Behavior depends on the actual npm version, project policy and selected command. Do not assume that every installation runs every dependency hook, or that a familiar suppression flag makes a later explicitly invoked test safe. This article did not install npm or execute those commands.

For Amara, that means an install-policy discussion does not answer what tests/check.cjs does. The explicit test remains a separate execution decision. Likewise, preventing an installation script would not independently stop an editor task or a development-container setup action. Map the action you intend to take to the configuration that actually controls it.

Avoid copying a proposed workaround into the terminal before understanding its effect. If suppressing setup would leave the exercise unusable, request the employer's approved alternative rather than silently changing the environment and submitting results from a different setup. An accurate account of what you could verify is more useful than a blanket claim that installation is safe.

An inspection tray separates configuration review from an execution path and an unsent clarification

A remote environment still needs a permission review

GitHub's Codespaces security documentation describes separate virtual machines and networks, while allowing outbound internet connections. It warns that devcontainer.json can install extensions and run arbitrary setup code, and identifies repository access, secrets and other granted features as security considerations. Moving execution away from a personal laptop does not erase the significance of those permissions.

A remote setup may be an appropriate employer-approved choice, but inspect what it would receive and run before creating it. Do not add personal credentials or broaden repository access merely because the environment is described as isolated. This review did not create a codespace or container, test isolation, forward a port or grant a token.

The relevant question is specific: which code would run, under which account, with what data and access? “It is in the cloud” does not answer that. Nor does an environment's successful creation establish that the recruiting request itself has been verified.

Worked stop decision: confirm a changed request through the existing channel

Fictional candidate Ben has a confirmed interview for a backend role. A later message from a different domain asks him to replace the agreed browser exercise with a ZIP project, run setup with administrator privileges, and provide his personal SSH private key to “enable the test.” These are stipulated fictional inputs, not a report about any named employer.

Ben completes a request-to-authority record:

Input Evidence available Completed decision
Original exercise Confirmed in the existing recruiting conversation Preserve it as the current agreed scope
New sender and ZIP Not confirmed through that conversation Do not download or run the replacement
Administrator setup No explanation tied to the original task Do not grant elevation while the request is unresolved
Personal private key Requested by the new message Do not provide the key or paste it into project configuration
Contact route Previously verified recruiting channel exists Prepare a clarification there, not at the new message's supplied address

His decision is stop before execution and verify the change. He does not need to run the setup to prove that the request warrants clarification, and he does not assert that the domain mismatch alone proves a scam.

His complete illustrative message remains unsent:

Our confirmed interview instructions describe a browser exercise. I received a separate request for a ZIP project, administrator setup and a personal SSH private key. I have not downloaded or executed the replacement or provided credentials. Could you confirm whether your team sent this change and provide the approved exercise and environment requirements through this existing channel?

Proceed only after the request and execution boundary are understood. Keep a record of the approved project, commands and unresolved limitations without copying secrets into it. As you find suitable roles on LandOffer, use the same review sequence for each technical handoff: verify the source, inspect the proposed actions and report what you actually reviewed.

Sources and Further Reading