Experimental and included in the v1.1.1 package; never part of the default scan path.
codeforge context . --query verification --limit 10 --max-bytes 32768
codeforge impact . --depth 4 --graphBoth commands print JSON, run locally, and read source without applying changes, starting analyzer tools or calling a model. Impact requires and invokes local Git to inspect changes. They reuse WorkspaceManager ignore rules, existing Tree-sitter grammars and Git status (staged, unstaged and untracked). They are not invoked by scan/review/Fleet or by the verification gate.
Context uses case-sensitive literal symbol-name matching. Each fragment carries
the relative source path, content hash, original symbol range, exact excerpt
range and a truncation flag. Line and character-column positions are one-based;
byte offsets are zero-based. A changed source hash invalidates the excerpt.
The entire workspace is enumerated using the existing scanner; parsing defaults
to 512 sorted files, with a hard maximum of 8192 files, 1 MiB per source, 20,000
symbols and 20,000 relations. Output is at most 100 fragments/256 KiB of source.
partial reports omitted, unsupported, invalid or changed source and bounds.
The graph contains function/method declarations and candidate call edges found by Tree-sitter. A matching callee name is heuristic, including when unique. All same-name targets are retained and ambiguous edges are labeled. Imports, scope/type resolution, callbacks, generated code, macros, dynamic dispatch and deleted symbols are not resolved. Cycles are visited once and traversal depth is bounded. Test naming/attributes supply suggestions, not a test selection guarantee. The full configured test/build/semantic gates are unchanged.
These commands do not provide whole-program static analysis or semantic verification. They expose evidence for agent inspection; they never convert a suggested relation or test into permission to apply a fix.