sean-review
Skill61 starsChanged 6 months agoApache-2.0View on GitHub
Review Ragu changes using Sean Bowe's priorities: protected contracts, minimal abstractions, and precise documentation. Use when Sean's perspective is requested.
In this repository15 other files it gives its agents.
- alex-reviewSkill
- book-fixmeSkill
- book-refineSkill
- book-reviewSkill
- clean-circuitsSkill
- code-refineSkill
- code-reviewSkill
- doc-auditSkill
- fv-docsSkill
- fv-preflightSkill
- fv-reviewSkill
- help-book-reviewSkill
- master-fvSkill
- spec-fidelity-auditSkill
- tal-reviewSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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
- Distinctive: line 14
Ask which states a correct caller can construct.
- Distinctive: line 35
Ask what concrete caller or generic bound justifies each added trait, wrapper, lifetime, macro, constant, or public method.
- Distinctive: line 38
Test access alone does not justify production API exposure.
- Distinctive: line 41
Keep layout and FFT details in the registry, strategy inside its crate, curve choices out of generic dependencies, and analysis state in …
- Distinctive: line 65
Prefer independent mathematical oracles and targeted failures over mirrored implementations.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.
Files
Sean review
Apply the relevant priorities below to the requested scope and exact revisions, using current code, affected callers, and repository policies. Give your own review, not Sean's approval. Keep feedback local unless publication is authorized.
Make the contract do the work
Ask which states a correct caller can construct. Put invariants in the API, type, token, builder, or guard that establishes them; do not accept a name or documentation claim as enforcement. Avoid scattered checks and caller bookkeeping. Unreachable error variants and repeated defensive checks can signal a missing abstraction. Distinguish protected internal assertions from fallible public inputs needing errors or Option. A private call site with a guaranteed invariant can unwrap; diagnostic debug assertions can be appropriate. Use typed retry conditions, not display strings, and rely on actual trait guarantees rather than one implementation's endianness or layout.
Check constraints rather than honest assignments
For gadgets, derive soundness and completeness from constraints, not honest witness assignments. Witness-only facts may need a separate type. Identify the designated stage enforcer and the binding that permits reuse; an unenforced stage is not automatically a bug. Context-free consistency APIs must re-establish constraints after prior context is stripped. Separate native advice cost from constrained arithmetic.
Keep abstractions small and responsibilities in place
Ask what concrete caller or generic bound justifies each added trait, wrapper, lifetime, macro, constant, or public method. Prefer inherent or existing operations over pass-through layers. Preserve domain structure instead of positional vectors. Test access alone does not justify production API exposure; prose cannot make a public item private.
Keep layout and FFT details in the registry, strategy inside its crate, curve choices out of generic dependencies, and analysis state in its owning driver. Share driver-generic algorithms where native and circuit behavior must agree; keep independently optimized evaluators separate when equivalence tests justify the boundary.
Find hidden coordination and misplaced state
Localize order-dependent indexing, synchronized counts, duplicated constraints, scope restoration on errors, and caller-carried layout advice. Ask for a concrete usage example before moving a burden from registration to every subsequent call. For routines, equal types, shapes, polynomial contributions, and reusable typed outputs are distinct obligations. Check input/local wire separation, output mapping, scope state, and nested deferred execution.
Require evidence proportionate to optimization complexity
Inspect allocations, normalization, inversions, eager work, and deep clones before adding machinery. Complicated parallelism and small constant-factor wins need representative benchmarks covering scheduling, caches, workload size, and higher-level parallelism. Straightforward simplifications and asymptotic gains need no arbitrary speedup threshold. Check enabled and disabled feature contracts.
Prefer independent mathematical oracles and targeted failures over mirrored implementations. Distinguish representation from mathematical equality; exercise changed constraints and driver side effects. Check required CI statuses when path filters skip work.
Make explanations precise and economical
Lead with the contract, then prerequisites and implementation details. Put protocol math, guide usage, and local rustdoc rationale at their proper layers; give each concept one authoritative explanation. Keep useful examples and subtle local rationale. Use rendered math, formatted identifiers, and API links in rustdoc. Check fixed/free variables, coefficient order, polynomial/vector operations, and assembly/evaluation. Describe current behavior, not the diff; avoid listing what rustdoc already renders or narrating obvious optimizations. Simplified guide excerpts must remain truthful; preserve safety and completeness preconditions.
Deliver a calibrated review
Identify the mismatch, its consequence, and the smallest coherent correction. Separate defects from preferences, future designs, and unverified questions; report current locations, scope, and checks actually performed. An incomplete improvement need not wait for an ideal redesign; there is no finding quota.
Consult references/evidence.md when provenance is requested or an ambiguity about applying a priority could change a finding. Historical examples do not override current code or repository policies.
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
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.

