miri

Bolt ⚡ — a performance-obsessed single pass that lands ONE small, measured, low-risk optimization in the Miri compiler/runtime. Profiles hot paths (parsing, type-checker traversals, MIR lowering, codegen, runtime intrinsics), picks the best clean win under ~50 lines, implements it with a comment explaining the optimization, measures before/after, and runs the full gate (make format/lint/build/test). Reads and appends critical learnings to .jules/bolt.md. Use for "make it faster", "find a perf win", "optimize this", or /miri-bolt [path]. If no clear win exists, stops without changing anything.

Scout 🔎 — a quality-focused single pass that hunts down ONE real bug in the Miri compiler/runtime and fixes it at the root. Reproduces the bug as a failing test FIRST, traces the full pipeline path to the confirmed root cause (file:line + evidence), fixes at the correct layer, and runs the full gate (make format/lint/build/test). Hunts crashes, wrong output, unhandled edge cases, missed error paths, unwrap/panic on hostile input, off-by-one, and exhaustiveness gaps. Reads and appends critical learnings to .jules/scout.md. Use for "find a bug", "hunt for bugs", "harden this", or /miri-scout [path]. If no real bug is found, stops without changing anything.

Sentinel 🛡️ — a security-focused single pass that finds and fixes ONE security issue (or lands ONE hardening enhancement) in the Miri compiler/runtime. Threat model is a compiler eating untrusted .mi source and a runtime handling untrusted data — hunts compiler-DoS (reachable unwrap/panic/index on hostile input), memory-safety holes (Perceus UAF/double-free, unsafe/FFI/ABI mismatch, buffer/bounds/GPU overruns), integer overflow, and path traversal in module/import resolution. Reproduces the issue, fixes it failing-securely, and runs the full gate (make format/lint/build/test). Reads and appends critical learnings to .jules/sentinel.md. Use for "security check", "harden this", "find a vulnerability", or /miri-sentinel [path]. Prioritizes ruthlessly — critical first. If nothing qualifies, lands one enhancement or stops.

CTO-orchestrated full validation pass over the current diff (default) or a specified location — fans out the specialist panel (Rust, Security, Software Architect, QA, Compiler Architect, GPU), then fixes all critical and major findings (and minor where cheap) via the Lead Miri Engineer, loops to green, and ends with a CTO verdict. Use for "audit", "review", "validate", "QA this", "security-check", "is this ready to ship", or /miri-audit. Replaces the old architecture-only audit and the retired miri-qa.

Full multi-agent panel execution of a Miri compiler task — CTO-orchestrated, spawns the architect + specialist subagent panel (Rust, Security, Software Architect, QA, Compiler Architect, GPU) and the Lead Miri Engineer. Use this ONLY when the full panel is explicitly wanted (high-risk Major-tier work, deep multi-perspective review, or when the user asks for "the panel"). For everyday features prefer the lighter single-agent `miri-task`. Classifies the risk tier (Trivial/Standard/Major per PRINCIPLES.md §8) and scales review to the tier. Loops until the CTO declares done with no open critical/major issues.

Fast single-agent end-to-end execution of a Miri compiler task — no subagents. You implement the feature yourself with TDD (Red-Green-Refactor), then run a self-review folding in every specialist lens (Rust idiom, Perceus/memory safety, architecture/SOLID, test coverage/honesty, compiler-design soundness, GPU), and finish with QA on your own work. Done ONLY when format, lint, build, the full test suite, and `make audit` are all green and the self-QA pass finds no open critical/major issue. Use for everyday features and fixes. For high-risk Major work or an explicit multi-perspective panel, use `miri-panel-task` instead.