Glossary

A per-PR-rebuilt structured table of entities × layers × invariants × tests × last-touched, generated automatically so that every Claude Code session opens against the same architectural snapshot of the codebase.

A posterior probability — updated from observed PostToolUse build and test outcomes — that a given plan step will succeed when executed, used by the agent to reorder or pause its plan in real time.

Inline `[VERIFIED]` / `[ASSUMED]` / `[MISSING]` markers attached to every symbol reference in a Claude Code plan, forcing the planning subagent to declare its evidence before any implementation begins.

A retrieval pattern combining classical BM25 lexical search with dense vector similarity, used by AI agents to recover both literal-match results and semantically-related passages from a codebase or knowledge base.

A Claude Code PreToolUse hook that blocks any tool call referencing a symbol the agent has not yet read into context — forcing the agent to investigate before it acts.

A Model Context Protocol server: a process that exposes tools, resources, and prompts to an LLM agent over a standardized JSON-RPC interface.

A Claude Code PostToolUse hook that detects edit-thrashing — the agent flipping the same lines back and forth across iterations — by hashing recent diffs in a sliding window and stopping the session when oscillation is detected.

A code-extracted representation of recurring patterns in a codebase — naming conventions, folder structure, modifier usage — that subagents read before generating new code so the output matches what a human reviewer expects.

A Claude Code PostToolUse hook that runs `dotnet build` and `dotnet test` after every multi-file edit, surfacing failures back to the agent before the next plan step so broken intermediate states never accumulate.

The compiler-grade symbol resolution Roslyn produces for C# code — type, binding, and reference information that AI agents can use as ground truth instead of guessing from text.

A long-running monitor process — distinct from a hook — that watches a Claude Code session for token-budget exhaustion, oscillation hashes, and plan drift, with the authority to inject a Stop event.

Cost-aware routing of Claude Code work to different model tiers — Haiku for searches, Sonnet for implementation, Opus for planning — so that token budget is spent on the steps where reasoning actually pays off.