Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
64 lines (52 loc) · 3.32 KB

ARCHITECTURE.md

File metadata and controls

64 lines (52 loc) · 3.32 KB
Raw

Architecture

CodeForge joins analysis and change verification through shared contracts while preserving language-native semantics. CLI and desktop call the same core; no background service, database or model is required for a CLI invocation.

Data flow

Repository → Workspace + Git selection → Language engines
                                         ↓
                                    Diagnostics → Baseline classification
                                         ↓
On-demand source context ───────────→ Transformation preview
                                         ↓
                              Verification + benchmark evidence
                                         ↓
                              Policy / transaction / report

Context extraction is explicitly requested, not a mandatory review stage. Benchmarking, external analyzers and configured build/test commands are separate operations. The hot/warm/cold scheduling categories organize work; they do not promise incremental Tree-sitter caching or semantic resolution.

Responsibilities

Crate Responsibility
codeforge-protocol Diagnostic, transformation, verification and SARIF data
codeforge-workspace, codeforge-git Enumeration, hashes, language detection and changed-file selection
codeforge-engines Native grammars, selected rules, tool discovery and adapters
codeforge-diagnostics, codeforge-baseline Aggregation, fingerprints and reviewed finding classification
codeforge-scheduler Task priority, cancellation and bounded scheduling
codeforge-transform Patch preview, transaction history, apply and undo
codeforge-verification, codeforge-benchmark Bounded process execution and measured results
codeforge-plugin Declarative metadata and manifest validation
codeforge-core Single-workspace orchestration and on-demand context
codeforge-fleet Repository policy, stable tasks, reports and optional PR publication

Trust and failure boundaries

A parser finding, imported path or suggested test has no application authority. Application policy evaluates explicit check results. Fleet adds stricter required-check and publication gates; a generic verification accepted flag alone is not equivalent to Fleet approval. See Verification.

Missing tools, malformed output and timeouts are reported as unavailable or failed work. Each Fleet repository retains its own transaction and outcome. Partial success is possible; there is no fleet-wide atomic rollback. Process limits are not OS isolation, as described in Security.

Persistence and evidence

Transactions retain original/transformed content and history under .codeforge/history/. Reports contain diagnostics, patches and verification results. Fleet adds run.json, summary.md, tasks.json and per-repository bundles under the configured report directory. Task/run/repository identifiers in existing tracing spans associate execution with evidence, without a new tracing service or dashboard. See Evidence Model.

The baseline records reviewed debt; it does not suppress evidence or prove a change safe. No Rust crate or runtime service is shared with LCE. Optional language tools and CodeQL remain external processes supplied by the user.