## TL;DR
Stop auto-deleting on checker output. Verify each flagged link with a real browser user-agent and retries; tune the checker for bot-blocking hosts. Require human approval before removing any link the checker flags.

## The exact error
```text
link checker false positives deleted working links
```

## Steps
1. **Restore the deleted links**
   Review the agent's recent link-removal commits and restore every link that is actually live.
   Success check: All falsely removed links are back.
2. **Verify with a real client**
   Re-check the flagged URLs with curl using a browser user-agent and retries; most false positives are bot-blocking, not dead links.
   Success check: You know which flags were false positives and why.
3. **Tune the checker**
   Set a browser user-agent, raise retries and timeouts, and exclude known bot-blocking hosts from hard failure.
   Success check: The checker no longer flags healthy bot-guarded pages.
4. **Gate deletions on review**
   Change the agent's policy: flagged links get reported, never auto-deleted; a human approves removals.
   Success check: No link is deleted without human confirmation.

## When to use
- Use when a link checker deletes links that are actually live.
- Use when an agent maintains docs links autonomously.

## When not to use
- Not for genuinely dead links (those should be fixed or removed).
- Not for build-time link validation (different tool, different policy).

## Why it happens
Many sites block non-browser user-agents or require JavaScript, so a naive checker reports them as broken. An agent that acts on the report without verification converts false positives into real damage.

## Edge cases
Deletion is the irreversible action; everything before it should be verification. Keeping a report of flagged-but-kept links lets the team tune the checker without losing the signal.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_quDE0qOmPtrc2c89ViC8Hg