Technical assessment for the AI era
SkillFoundry is a repo-based, AI-judgment-scored assessment. Candidates work a ticket inside a real repository with an AI pair engineer, and the score records the judgment they applied to that pair, with an evidence trail the hiring team can review.
How an assessment runs
Assessment mechanics
A candidate opens a repository, reads a ticket, and works the change with an AI pair engineer. Scoring dimensions describe that work. The direction and verification scoring model describes the judgment they applied to the pair.
Real-codebase ticket-style tasks
Candidates receive a ticket in a real codebase, with tests and acceptance criteria attached. The work is the kind of change a team would put on a board. It is not an algorithm puzzle, a whiteboard drill, or a toy problem stripped of context.
AI pair engineer
An AI pair engineer sits in the workspace and can propose code. The candidate decides what to ask for, what to keep, and what to reject. The pair is part of the task, the way a copilot is part of the job.
Scoring dimensions
The report describes how the candidate carried the ticket, not only whether the tests went green. The scoring and reporting section names every dimension from the published sample behavioral report.
Direction and verification
The direction and verification scoring model measures the judgment applied to the pair. Direction is how the candidate framed the work. Verification is how they checked what came back before it shipped.
Evidence-based assessment
See how the work was done, backed by evidence
SkillFoundry looks past whether the tests pass to how a candidate investigated, used AI and verified their work. Reviewers can check every result against the evidence.
How they investigate
Whether the candidate took the time to understand the problem and the code before changing it.
How they use AI
AI use is not penalized. What matters is the judgment a candidate applies to what the AI produces.
How they verify
Whether the candidate checked that their solution works, alongside the result of grading the code they submitted.
SkillFoundry is live. Evidence-based results become the primary score after a readiness review.
What reviewers receive
Scoring & reporting
AI judgment is measured from the work a candidate does with an AI pair engineer on a ticket inside a real repository. The pair can propose code. The result does not rise or fall because the candidate used the pair. Copilot Command scores whether they directed that output and whether they verified it. Direction is the framing they gave the pair: the ticket constraint, the part of the codebase in scope, and the change they asked for. Verification is what they did before the proposal became the submission. They read the diff, ran the task tests, and rejected output that failed those checks. A candidate who accepts generated code without checking it is not scored the same way as a candidate who steers the pair and proves the result.
A reviewer reads that judgment on the decision brief, which shows Outcome, Investigation, AI judgment, and Verification. Outcome is the submitted code against the ticket. Investigation is how the candidate read the repository before editing. AI judgment is the direction they gave the pair. Verification is the checking they did before calling the work done. Each area points at the evidence trail for that candidate. A hire or no-hire recommendation is attached to the trail with a confidence note, so a reviewer can see how complete the record was. When a signal was not captured, the report says the evidence is missing. Missing evidence is never counted against a candidate.
Evidence trail
Every candidate has an evidence trail: the ticket, the submitted code, the test results, and the record of how they worked with the pair. Reviewers check the recommendation against that trail. The original result stays available when a person adjusts it, and the adjustment is logged.
Hire or no-hire recommendation
The report offers a hire or no-hire recommendation with a confidence note tied to how complete the evidence is. The recommendation is for the reviewer. It does not replace the hiring decision, and a gap in the record is stated rather than filled in.
Copilot Command
Copilot Command scores whether the candidate directed and verified AI output rather than whether they used AI. Using the pair is allowed wherever the organization permits it. The score is the direction and the verification.
Behavioral dimensions
The published sample behavioral report names these dimensions. SkillFoundry lists them here without sample scores.
- Task Outcome & Code Quality
- Problem Solving & Strategy
- Execution Discipline
- Collaboration & Communication
- Requirements & Product Thinking
- Risk Management & Safety
- Velocity & Time Management
- Tool Mastery
- Learning Behavior
- Professionalism & Reliability
Decision brief
On the current decision brief, a reviewer sees Outcome, Investigation, AI judgment, and Verification.
- Outcome. The submitted code, read against the ticket and the tests that ran on it.
- Investigation. How the candidate read the repository and the ticket before editing.
- AI judgment. The direction they gave the AI pair engineer, and whether that direction matched the work.
- Verification. The checks they ran on AI output and on their own changes before submitting.
For the person taking it
Candidate experience
Browser and VS Code
Candidates take the assessment in the browser or in VS Code. Both open the same ticket and the same repository. The task does not change with the surface.
What candidates see
They see the ticket, the codebase, the tests, and the AI pair engineer when the organization allows it. Before the session they are told what is captured. Afterward they receive descriptive feedback on how they approached the work.
Missing evidence is never counted against a candidate.
For the organization
Enterprise capabilities
- macOS and Windows desktop agents. Native agents for assessments that need a local environment, and for consented workforce intelligence on engineers already hired.
- Workforce intelligence. The same evidence model, applied with consent to the team you already have, so coaching starts from recorded work. Workforce intelligence is Enterprise.
- Team benchmarking. Baselines come from assessments completed by engineers already on the hiring team, so the comparison is that organization's own bar. Team benchmarking is Enterprise.
- SSO/SCIM. SAML single sign-on and SCIM provisioning for the people who administer and review assessments.
- ATS integrations. Connectors that start assessments from the applicant tracking system and return the result, including Greenhouse, Ashby, and Workable. ATS connectors are Enterprise.
- Custom branding. Your logo, colors, and product name on the candidate experience. Custom branding and white-label are Enterprise.
What you can defend
Compliance
Hiring tools that score people need a posture a counsel team can read. SkillFoundry states that posture in plain language.
NYC Local Law 144 bias audit tooling is built in, so a customer can support the bias-audit work that law expects of automated employment decision tools.
EU AI Act transparency is documented: what the assessment measures, what a reviewer sees, and what the product does not claim to know.
The SOC 2 Type II examination is in progress, targeting Q4 2026. GDPR and CCPA data-subject rights are supported. A DPA is available.
Start with a ticket your team can defend
Run a repo-based assessment, read the evidence trail, and decide whether the record is one your legal and hiring teams can share.