fix: prevent fetch loops with call tracking and caching (#26)

[vtemian]

Summary

Fixes #26 — agents entering infinite loops fetching the same web documents repeatedly.

• Re-enable doom loop detection — doom_loop: "allow" was set as collateral in e4c2f47 when all permissions were blanket-allowed. Removing it restores OpenCode's built-in loop protection.
• Add fetch tracker hook — per-URL call counting with LRU-cached results. Warns after 3 repeated fetches, hard-blocks after 5. Tracks webfetch, context7_query-docs, context7_resolve-library-id, btca_ask.
• Session-scoped — cache and counts are cleaned up on session delete, following existing patterns.

26 new tests, all passing. No changes to existing test results.

[cubic-dev-ai]

1 issue found across 4 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/hooks/fetch-tracker.ts">

<violation number="1" location="src/hooks/fetch-tracker.ts:190">
P1: Max-call blocking is gated by cache-hit conditions, so repeated fetches can bypass the hard block when cache is expired/missing.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}