kodeatlas.resolve_code_anchor
Resolve a historical or textual code reference to the current symbol version and report confidence.
Agent-native context · In development
Give the agent the evidence around the code—not an unbounded history dump.
KodeAtlas will not be the coding agent. It will be the trusted memory, context, and coordination service that coding agents query before and during work.
Target workflow
Resolve, contextualize, claim, refresh, and record.
- Resolve task references to symbols, tests, endpoints, and code anchors.
- Return a context packet constrained by relevance and token budgets.
- Record assumptions and intended semantic write scope before editing.
- Warn when related symbols or contracts are already being changed.
- Refresh context when the base commit or dependencies change.
- Record actual changes, tests, outcomes, decisions, and released claims.
Proposed MCP tools · Future surface
One interoperable protocol, with CLI and HTTP equivalents.
These tools describe the intended Phase 1–4 interface. They are not part of KodeAtlas CLI v0.2.0 today.
kodeatlas.get_symbol_context
Return a bounded context packet for a symbol at a commit or branch.
kodeatlas.get_change_history
Return lineage, changesets, pull requests, issues, reviews, and decisions relevant to a symbol.
kodeatlas.get_change_impact
Return callers, callees, tests, historically coupled code, planned work, and active claims.
kodeatlas.get_regression_history
Return evidence-scored defects, reverts, incidents, test failures, and corrective work.
kodeatlas.get_related_backlog
Return open or recently completed work connected to the symbol or its dependencies.
kodeatlas.claim_work
Create a time-bounded semantic claim for an agent and task.
kodeatlas.check_conflicts
Evaluate proposed or active work against other claims and current repository state.
Bounded packets
Deterministic for the same evidence snapshot and request parameters.
A context packet is intended to include:
- repository and commit identity;
- the resolved target and current structure;
- historical changes and related decisions;
- risk signals and prior regression evidence;
- planned work and active claims;
- confidence, completeness, and source pointers;
- generation version and token budget.
Raw evidence remains separately queryable.
Coordination · Planned
Semantic intent is more useful than blunt file locking.
Two agents may safely edit independent functions in the same file. Two other agents may alter the same API contract from different files. KodeAtlas is designed to track symbols, dependencies, tests, schemas, endpoints, branches, worktrees, expected contract changes, heartbeats, and expiring claims instead of treating a file as the only conflict boundary.
Take the next step
Test agent context against a repository with real history.
The Phase 1 question is simple: does an evidence-backed context packet materially improve the agent’s implementation plan?