Evaluating a Proposed Fix
The families of fixes
Once a CVE is confirmed as genuinely exposed, several strategies are available:
- Version upgrade — the simplest, when a fixed version exists and stays compatible.
- Targeted override (forcing a sub-dependency to a patched version, without touching the parent package) — useful when the direct dependency hasn't published a fix yet.
- Workaround — disabling or replacing the vulnerable feature if it isn't essential.
What to check before accepting
A fix that "compiles" isn't necessarily a good fix — three questions are worth asking every time:
- Breaking changes — does the upgrade change an API used elsewhere in the project?
- Actual scope of the fix — does it really address the code path identified as vulnerable, or only part of the problem?
- Relative risk — is the risk introduced by the change (regression, incompatibility) lower than the risk of leaving the vulnerability as-is?
In the vast majority of cases, the simplest answer — a minor or patch version bump — is enough and breaks nothing. But that's still something to check, not a default assumption: a major upgrade may well call for a broader code review before it's accepted.
The fix's role in the mission
The fix proposed by the agent is only a proposal, never an execution. The decision to apply it — and the responsibility for that decision — stays human. The next module covers the guardrails around all of this.
Check your understanding
What kinds of fixes can an agent propose for a confirmed CVE?
Why shouldn't you automatically accept the first fix the agent proposes?