agentleFS
tachyon-zcashtachyon-zcash/ragu/.claude/skills/tal-review/SKILL.md

tal-review

Skill61 starsChanged 6 months agoApache-2.0View on GitHub

Review Ragu changes using Tal Derei's priorities: end-to-end verification, ordering, meaningful tests, and API boundaries. Use when Tal's perspective is requested.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    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:

    Use history before dismissing a removed check or test.

    line 30
  2. Distinctive:

    Check driver remapping, scope restoration, staging, and error returns.

    line 45
  3. Distinctive:

    Put shared logic in the layer that owns it without reversing dependencies.

    line 53
  4. Distinctive:

    Follow allocations, deep clones, borrowing, reference-counted sharing, copy-on-write, caching, and repeated commitments through the hot …

    line 61
  5. Distinctive:

    Check book pages, examples, comments, tests, and benches for stale contracts.

    line 67
  6. Distinctive:

    Start with motivation and concrete flow.

    line 70

≈ “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.

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

Tal review

Apply the relevant priorities below to the requested scope and exact revisions, using current code, affected callers, tests, and repository policies. Give your own review, not Tal's approval. Keep feedback local unless publication is authorized.

Follow the claim through the implementation

Trace what is produced, encoded, committed, stored, recomputed, and checked. Name where each value becomes authoritative. The native prover, circuit, verifier, and public instance must describe the same computation; a proof field or application claim is not verified merely by being present.

For recursion, separate the current step, child proofs, and deferred checks; track curves and fields. Explain transitive binding through commitments and equalities rather than honest witness generation.

Treat order and coverage as invariants

Inspect serialization, reversal, Horner powers, polynomial queries, claim iteration, stage registration, and commitment traversal together. Equal counts do not establish order or coverage. Prefer one authoritative enumeration or construction where it prevents drift, including special fields outside enums. Use history before dismissing a removed check or test; remove checks that are actually implied by surrounding invariants.

Ask tests to exercise the claimed behavior

Tests must reach the changed production path. Witness construction does not exercise synthesis, smoke tests do not cover every stage, and trivial proofs can make corruption tests vacuous. Use asymmetric dimensions, meaningful zero and one cases, success/error controls, and Known/Unknown combinations in nested routines. For binding tests, mutate a relevant component of a valid nontrivial baseline and state what rejection demonstrates. Finite tests are not soundness proofs; extend an existing property test when it covers the invariant directly.

Preserve contracts across boundaries

Check driver remapping, scope restoration, staging, and error returns. Gadget contracts concern wires, not witness values; context-free consistency must work after prior constraints are stripped. Distinguish a contract violation from a reachable exploit. Defensive checks need a real obligation and must not reject valid cases or duplicate existing enforcement.

Reduce friction without moving it to callers

Put shared logic in the layer that owns it without reversing dependencies. Challenge unused public methods, test-only production APIs, and traits or crates without consumers. Use semantic types or helpers when they simplify real calls. Assess refactors from the application developer's side: remove bookkeeping without moving internal distinctions to callers. Keep unrelated redesign separate.

Account for actual work and actual documentation

Follow allocations, deep clones, borrowing, reference-counted sharing, copy-on-write, caching, and repeated commitments through the hot path. Wrapping a clone in Arc does not remove it. Match performance claims to the actual workload and baseline, including full fuse/proof generation when claimed; distinguish operation counts from timings.

Check book pages, examples, comments, tests, and benches for stale contracts. Describe current behavior and label proposals. Distinguish encoding from verification, native arithmetic from constraints, and components from stages. Start with motivation and concrete flow; avoid duplicate implementations and fragile numeric inventories.

Give feedback in Tal's manner

Lead with the concrete concern and cross-file consequence. For uncertainty, ask the specific question and identify the check or change that would settle it. Cite prior discussions only for necessary rationale. Separate defects and demonstrated contract violations from questions, suggestions, and nits. Do not turn an intuitive concern into a confirmed bug. Report current locations, checks, and coverage limits; local validation alone does not establish merge readiness. 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, protocol, or repository policies.

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.