yuridevx/yd-skillset (algo-craft)

In-place algorithmic work on a narrow target — add a new algorithm, improve an existing one, refactor to the same intent, or evaluate that it should be left alone, scoped to a function, class, or single file. Twin creeds — performance does not exist until a benchmark says it does; correctness does not exist until it survives adversarial refutation. Linear spine — SCOPE (job, contract with guarantees and sync contract, correctness-lens risk scan, tags HOT/COLD CONCURRENT TRUST-BOUNDARY SUSPECT) → PIN (pin the contract before touching anything — PINNED/FLEXIBLE/SUSPECT behavior split, differential + property harness against the current code while it still exists, SUSPECT behaviors adjudicated so a bug is never enshrined as expected) → BASELINE (measured noise — interleaved repetitions with recorded spread, realistic plus worst-case shapes, contended when concurrent, allocation counts, machine profile) → DESIGN (at most 3 candidates, refuted on paper) → IMPLEMENT (in place, red-first, no dual paths) → VERIFY (risk-proportional adversarial ladder, hard cap 2 blind agents, "inline" in the request drops to zero agents with oracles unchanged — edge-hunter on hot or complexity-adding results, race oracle + interleaving-hunter on concurrent ones, the complexity class assigned by whichever hunter runs, boundary enforcement at trust boundaries, zero agents on cold simple paths) → MEASURE (adoption by proportionality; improve reverts on a miss, refactor adopts only if intent is preserved AND nothing regresses AND the code got simpler) → REPORT. Hot paths obey realtime rules — zero steady-state allocations, no hidden allocators, bounded worst case, no blocking. Testing method loads from the test-practice skill; language specifics from a <lang>-toolkit skill (e.g. msvcpp-toolkit) when one exists. Edits land in the existing files; no research fan-out, no parallel implementations. Trigger on "algo-craft", "/algo-craft", "add an algorithm", "implement an algorithm here", "improve th

Claude Skills are markdown + scripts loaded into Claude Code; scanned with the same pipeline.

Source:

github.com/yuridevx/yd-skillset

Provenance:

Found by crawling (provenance, not a security rating)

We found this by crawling public sources and nobody has attested to it. That is how most of the registry gets here, and it is not a mark against the artifact — it only means no authenticated act of publication is on record. An artifact can be the vendor's own and still appear here.