Skill Index — July 15, 2026

find-skills: discovery is only the first decision

The July edition placed find-skills at the entrance to the Skill ecosystem: describe a need, search a directory, then inspect candidates. It adds a discovery workflow rather than the capability you ultimately want. Finding a video tool, for example, does not itself generate a video.

Popularity can help narrow a search, but it answers a different question from suitability. Installation counts do not show continued use, and repository stars may cover many Skills rather than the specific one under consideration. The practical handoff is a shortlist with source links, followed by reading each Skill’s instructions, dependencies, and intended write access before installation.

ai-video-generation: the service remains a separate dependency

The archived review describes instructions for calling video models through inference.sh. Installing those instructions does not install the models locally or make generation free. The account, external service, and generation charges remain separate parts of the workflow.

That distinction changes how to assess the tool. A recurring batch workflow may benefit from shared job submission and file handling. An occasional clip may not justify another command-line login and service dependency. Before a batch, establish the chosen service’s current terms and cost, try one bounded job, and inspect the output. A completed job is not evidence that every generated clip is usable.

grill-me: make the unresolved decisions visible

The July review describes a requirements interview that examines the conversation and project before asking follow-up questions. Its useful output is a set of explicit decisions, not a longer specification filled with assumptions.

Consider the review’s habit-app example: reminders, offline behavior, AI failures, and time-zone boundaries all need decisions before implementation. An interview can expose those gaps, but can also expand indefinitely. Limit it to the next deliverable, record unknowns as unknowns, and stop when the acceptance conditions are clear. This example explains the workflow; it is not a newly measured productivity result.

How to use this archive

These are condensed readings of the site’s July 15 reviews, not new hands-on tests. Historical popularity and audit badges are not current guarantees. The former numerical scores were withdrawn because no reproducible scoring record was supplied. This English edition instead separates workflow descriptions from unverified outcomes. Recheck the linked source instructions before using any package.