Testing: web agent pipeline

[functor-flow]

Testing: web agent pipeline

full spec at spec

Status: Very limited.

Role. Provide module-level and end-to-end tests for the new web-search agent pipeline. This is an essential part of the system: the tests must run the pipeline on real Validator data and include cost benchmarking.

Responsibilities.

• Add focused tests for each boundary (Input, Serper, Crawler, Embeddings, Nav, SearchHistory, Decision).
• Add end-to-end tests hitting the agent entrypoint on real historical predictions from the Validator and checking both outcomes and evidence, while recording cost metrics.
• Make it easy to compare cost across runs (LLM tokens, SERP/Scraper usage) so budget and behavior regressions are visible.
• Document key findings from the testing process (edge cases, failure patterns, cost surprises) and share them with the team in Discord so the pipeline can be tuned collaboratively.

---

TODO

• Add small, focused tests around each module boundary so SearchTask → SearchHistory transitions remain stable as the pipeline evolves.
• Add at least one end-to-end test that replays a small, fixed set of Validator predictions through the agent entrypoint and asserts on the resulting AgentResult shape, key evidence fields, and basic cost metrics (without depending on live APIs).

[linear]

SWARM-72