Your team is putting its expertise into AI skills and workflows. Mensova helps you see what needs attention, which changes are helping, and what deserves wider use.
Team members can find shared skills, submit feedback, and request improvement suggestions. Managers can review performance trends and security findings, start configured benchmark and AutoImprove runs, and approve improvement suggestions, with the appropriate permissions.
Developers can bring feedback, skill context, and status checks into terminal workflows, or connect Mensova to their own applications and processes.
Detailed AutoImprove candidate review and export, along with Comparison, take place in Mensova’s web workspace.
Which skills need a closer look before more people start using them?
Mensova checks skill content for potential risks, including prompt injection, suspicious instructions, and unsafe handling of sensitive information. Findings show what was flagged, why it matters, and a recommended response.
Your team reviews the findings and decides what to change. You get a clearer basis for prioritising security work, with evidence to guide the conversation.
Checks make potential risks visible; they cannot guarantee that every threat has been found.
Linked note tells the agent to ignore earlier instructions PROMPT_INJECTION
Summaries include full customer contact details SENSITIVE_DATA
Bash granted alongside Write/Edit tools PRIVILEGE_ESCALATION
A convincing answer isn’t always a useful one. Does the skill actually help your team get better results?
Start with a generated benchmark draft, import existing tests, or write your own. Your team reviews the tasks, expected outputs, and scoring criteria before saving and running them.
Compare performance against no skill or an earlier version, using the same cases and evaluation settings. Inspect the answers, scores, and reasoning, and see where the evidence remains uncertain.
You get a clearer basis for deciding whether a skill earns its place, and whether a change is worth keeping.
| Criterion | No skill | v1 | v2 |
|---|---|---|---|
| Accuracy | 3.3 | 4.0 | 4.2 |
| Sourcing | 1.6 | 2.9 | 3.1 |
| Useful summary | 3.2 | 3.6 | 3.8 |
| Overall | 54% | 70% | 74% |
Your team has spotted a weakness. Which change is worth taking forward?
AutoImprove generates alternative versions of a skill and tests them alongside the unchanged original, using the same frozen benchmark. The team can review the complete instructions, score differences, and statistical evidence.
The person responsible can approve a candidate for export and manual application, or keep the original.
You can see what changed and what evidence supports the choice. That makes improvement work easier to review and prioritise, without assuming every experiment will produce a better skill.
| Candidate | Score | Delta |
|---|---|---|
| Original (unchanged) | 74.0% | 0.0 |
| Requires a source for every financial figure | 81.5% | +7.5 |
| Restructures summaries into decision sections | 76.2% | +2.2 |
| Shortens instructions and removes examples | 70.4% | −3.6 |
Several people have built skills for the same job. Which approach should the team rely on?
Comparison assesses two to seven skills against shared cases and agreed quality criteria. Results show how the candidates perform, whether they meet the required standard, and whether the evidence supports a meaningful preference.
An authorised manager can designate an existing skill as the standard for a defined task and scope, or record that the decision needs to wait.
The decision history preserves the rationale, evidence, and exact skill version. Alternatives remain available.
You get a defensible way to choose a shared standard, and revisit that choice as the work evolves.
| Criterion | account-research | account-brief | prospect-notes |
|---|---|---|---|
| Sourcing | 4.4 | 2.8 | 3.0 |
| Summary clarity | 3.2 | 4.5 | 3.1 |
| Accuracy | 4.1 | 4.0 | 3.6 |
| Overall | 3.9 | 3.8 | 3.2 |
| Meets standard | Yes | Yes | No |
Several people have built skills for researching customer accounts.
Flag potential risks for review.
Test sourcing, accuracy, and useful summaries, including whether a skill adds value over no skill.
Test alternative instructions for a skill that needs work.
Assess the available approaches against shared criteria.
Know where your team’s AI skills need work. See which changes hold up in testing. Build shared standards around evidence.
Competence compounds. Mensova keeps the record.