Skills
A dedicated section for agent skills: what they do, what pain they remove, and how to actually use them — written from the source, not the announcement.
grill-me — the skill that refuses to build until you can defend your plan
An interview skill that walks your plan down every branch of the decision tree — one multiple-choice question at a time — until there is no ambiguity left to discover mid-implementation.
By RobMitt ·490 stars ·Updated 2026-09-10handoff — the skill that survives context compaction and chain-links your sessions
A session-handoff skill that mines your conversation for decisions, failed approaches and measurements, then writes a structured, self-validating brief a fresh session can resume from — without repeating dead ends.
By REMvisual ·49 stars ·Updated 2026-09-10
What a skill is, and how to judge one
A skill is a markdown file of instructions plus a description the model uses to decide when to load it. There is no runtime and no code path; what gets loaded is the prose itself. That means a badly written skill fails the way a badly written prompt fails — the model reads it and does the thing you had not thought through, confidently.
This is a different thing from an MCP server, and the two are often confused because both are described as "extending" an agent. An MCP server gives the model capabilities it did not have: a tool to call, a data source to read. A skill gives it a method — when to do something, in what order, and under what constraints. Installing an MCP server changes what your agent can do. Installing a skill changes how it does what it already could.
The useful thing to read before installing one is the skill body, not the README. Star counts tell you that people found it worth bookmarking; they tell you nothing about what it will do to your workflow. Read the description field with particular care, because that is the trigger: the broader it is written, the more often the skill will insert itself into tasks you did not intend it for. A skill that fires on "any code change" is a different proposition from one that fires when you explicitly ask for a review.
Entries here are written after reading the skill source. Where a skill makes a claim about what it prevents, the entry says whether the instructions actually contain the check, because the gap between the two is where most of the disappointment lives.