[ajianaz]

Summary

Migrate engine layer from anyhow to typed CoraError enum via thiserror = "2".

New file:

• src/error.rs — CoraError enum with 17 variants

Modified (10 files):

• Cargo.toml — added thiserror = "2"
• src/main.rs — added mod error
• engine/llm.rs, engine/review.rs, engine/scanner.rs, engine/cache.rs → Result<T, CoraError>
• config/loader.rs, config/schema.rs → typed errors
• git/diff.rs → typed errors
• commands/review.rs → map_err(Into::into) for CoraError→anyhow conversion

CoraError variants: LlmRequest, LlmStatus, LlmParse, LlmStream, LlmJsonRepair, ConfigParse, ConfigRead, NoApiKey, DiffTooLarge, InvalidRef, GitCommand, NotInGitRepo, HeadDetached, CacheIo, CacheSerialize, ScannerError, AuthError, HookError, Timeout

Design decisions:

• Engine layer → Result<T, CoraError> (typed, programmatic matching)
• CLI layer → retains anyhow::Result (ergonomic, auto-converts via ?)
• #[from] for reqwest::Error and std::io::Error (standard thiserror pattern)

Verification: 224 tests ✅ | clippy clean ✅ | fmt clean ✅ | local binary test ✅

Prerequisite for: #116 (pre-LLM rule engine needs typed error output)

⚠️ Cora pre-commit found 2 false positives on #[from] attributes — these are valid thiserror patterns (different types per variant, no conflict).

Closes #86

Co-Authored-By: codecoradev noreply@github.com

[github-actions]

🔍 Cora AI Code Review

❌ Blocked — critical issues found.

🔴 Error (2)

• src/error.rs:72 — The #[from] std::io::Error on CacheIo generates a blanket From<std::io::Error> for CoraError. This means any function returning Result<T, CoraError> that calls something returning std::io::Error will silently convert the IO error into CacheIo, even when the IO error is unrelated to caching (e.g., config file reads, git operations). This can lead to misleading error messages. For example, std::env::current_dir() in open_repo() returns std::io::Error, and with ? it would be converted to CacheIo instead of a more appropriate variant.
• src/error.rs:15 — The #[from] reqwest::Error on LlmRequest generates From<reqwest::Error> for CoraError. Since reqwest::Error can represent both request failures and response body read failures, this blanket conversion means any reqwest::Error will always become LlmRequest, even when it might be more appropriate as LlmStream or another variant. Additionally, this can conflict if other variants also try to use #[from] with overlapping error types.

---

Review powered by cora-cli · BYOK · MIT