agentleFS
JuanMarchettoJuanMarchetto/agent-skills/skills/noricum-dev/SKILL.md

noricum-dev

Skill5 starsChanged 8 months agoMITView on GitHub

Development conventions for the Noricum C-to-Rust migration agent. Architecture decisions, crate structure, coding standards, and contribution guidelines. Use when: contributing to Noricum, noricum development, noricum architecture.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every migration must pass differential testing (byte-exact stdout match)

    line 81

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Noricum Development Conventions

Architecture

  • Noricum is an agent orchestrator, not a compiler
  • C2Rust is "step zero" (subprocess), not reinvented — but often skippable (P4: --skip-c2rust)
  • LLM agents are central from v0
  • Semantic Code Map (noricum-ir) tracks metadata, not compiler IR
  • rig-rs 0.31 for LLM integration (Rust-native, rustls)

Crate Dependencies (layered)

noricum-cli -> noricum-core -> noricum-agents -> noricum-tools -> noricum-ir
                            -> noricum-validation -> noricum-tools -> noricum-ir
noricum-mcp -> noricum-core

Coding Standards

  • Edition 2024, thiserror for lib errors, anyhow for CLI
  • tracing for logging, never println! in library code
  • No unwrap() in library code
  • Async with tokio
  • Tests in #[cfg(test)] mod tests within each file
  • C compiler: -std=gnu11 (not c11, needed for POSIX like strdup), -lm for math

State Machine

Pending -> Extracted -> Characterized -> C2RustDone -> Analyzed -> Refined -> Validated
                                                                    |
                                                            Repairing (max iterations)
                                                                    |
                                                            FallbackUnsafe

Model Router (Claude 4.6 era)

  • Easy: claude-haiku-4-5 (fast, cheap)
  • Medium: claude-sonnet-4-6 (also used for analysis)
  • Hard: claude-opus-4-6 (translation, complex repair)
  • Fallback: Ollama qwen2.5-coder:32b when no API key

Pipeline Improvements (P0-P5, learned from miniz + genann migrations)

  • P0: Quality floor — repair rejected if unsafe count exceeds translation baseline
  • P1: Best-version tracking — keeps highest-score version, uses it for fallback instead of c2rust
  • P2: Per-function C2Rust context — extracts only matching c2rust functions per chunk
  • P3: Incremental per-module — split_into_modules() groups C functions by prefix
  • P4: Skip C2Rust — --skip-c2rust flag; LLM often translates better without c2rust noise
  • P5: Idiomatic improvement hints — when code compiles + diff passes but score < threshold, generates actionable refactoring hints (reduce as casts, use iterators) so repair agent improves style instead of returning unchanged

Key Thresholds

  • Chunked translation: >800 LOC (MEDIUM_FILE_LOC)
  • Reduced repair iterations: >1000 LOC
  • Very large file handling: >2000 LOC
  • Structural summary: >800 LOC
  • Quality gate: re-translate if >5 unsafe blocks
  • Stall detection: 2 consecutive unchanged error counts → re-translate at temp 0.7
  • Chunk targets: 400 LOC (medium), 500 LOC (very large)

Pipeline Improvements (P30-P33, learned from miniz_zip.c 9 runs)

  • P30: Hybrid Repair Engine — 3-phase: rule engine (free) → surgical per-function (cheap) → legacy whole-file (expensive)
  • P31: Assembly Cleanup — fence stripping, syntax error parsing, use import merging
  • P32: Brace-Balance Validation — detect/fix truncated module outputs before assembly
  • P32b: Smart Truncate + Re-translate — truncate at last balanced brace, re-translate truncated modules
  • P33: Type Contract — types-first modular migration:
    • generate_type_contract() in type_contract.rs — single LLM call before module translation
    • ModuleSplit struct returns shared_context from split_into_modules()
    • resolve_header_types() reads #include'd .h files for complete type definitions
    • Assembly seeds P27 dedup from contract type names
    • Key learning: shared_context only has .c file content; .h headers must be resolved separately

Key Patterns

  • Functions migrate independently, ordered by dependency graph (topological sort)
  • Every migration must pass differential testing (byte-exact stdout match)
  • Translation cache: .noricum-cache/ keyed by SHA-256 of C source
  • RAG patterns: successful migrations auto-added to PatternStore for future context
  • Structural chunking: data model (structs/constructors) in chunk 0, logic functions in later chunks
  • For multi-file C projects: type definitions live in .h headers, not .c files — must resolve includes
  • Type contract prompt must enforce idiomatic Rust (no raw pointers, no C-style aliases, complete structs)

Powered by Noricum — autonomous C/C++ to Rust migration agent

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.