Claude Code now demonstrates that a fresh session can discover a complete component graph and shared contract definitions by reading only the immutable dependency records emitted by the Bit Cloud build platform. The change matters because it reduces the reliance on manually maintained memory files while also exposing the limits of record‑driven reasoning when behavior and runtime guarantees are required.
Dependency Records as a Source of Context
When a new Claude Code session starts with an empty workspace, the agent invokes the Model Context Protocol (read_scope) to fetch the public scope bit-oss.support. The response lists the four versioned components—React front‑end, Express service, shared ticket entity, and gateway platform—and their declared dependencies. Subsequent read_components calls return API references and file inventories, allowing Claude to bit import the source, analyze it, and apply a requested change (client‑side ticket filtering) without any prior conversation history or local code.
Limits of Record‑Based Reasoning
Records stop at structural facts. The API reference documents routes such as GET /tickets and POST /tickets/:id/comments, but it does not confirm that the service implementation actually supports those routes. The test suite reports a green build, yet integration tests that require a running MongoDB instance are skipped when the CI runner cannot start the database. Consequently, a passing build does not guarantee that the live service behavior matches the recorded contract.
Operational and Security Considerations
Relying on dependency records for initial code generation shifts the verification burden to downstream testing and manual validation. Practitioners should be aware that:
- Immutable versioned records provide a reliable snapshot of declared dependencies, but they lack runtime semantics.
- Memory files such as
CLAUDE.mdorAGENTS.mdcapture intent, timestamps, and decision rationale, which are not tied to a specific code version. - Missing or skipped integration tests can mask mismatches between documented contracts and actual service behavior, increasing the risk of runtime errors.
- Manual persistence checks (e.g., creating a ticket, adding a comment, restarting the platform) remain necessary to confirm data durability.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Engineers should treat dependency records as a strong foundation for structural discovery but supplement them with explicit behavior verification. Incorporate comprehensive integration tests that exercise declared routes, ensure CI environments can reliably start required services, and maintain concise memory notes for intent‑level decisions such as release dates or stakeholder contacts. By pairing immutable records with purposeful testing and intent documentation, teams can leverage Claude Code’s rapid onboarding while preserving operational confidence.