Jira at 9.5: Anatomy of the Best Skill in the Batch
The highest-scoring skill in the new batch is a tool integration: jira at 9.5. That is not the usual pattern, which makes it worth reading closely.
The six dimensions:
| Dimension | Score |
|---|---|
| Trigger | 9.5 |
| Structure | 9.0 |
| Workflow | 9.5 |
| Content | 9.5 |
| Engineering | 9.0 |
| Security | 10.0 |
| Weighted total | 9.5 |
Four properties explain most of the score.
1. It starts with capability discovery
The skill does not assume the agent knows which Jira backend is available. It instructs the agent to discover whether it is operating through a CLI or an authenticated MCP connector, and to read the relevant reference file only after that choice is made. That is a workflow decision, not an API list.
2. It routes between backends explicitly
The file names the two paths and the conditions for choosing each. An agent that has both a CLI and a connector does not have to guess. The routing is written down.
3. It gates mutations
Simple reads require capability discovery but not approval. Writes require a mutation approval step. The skill states the distinction and makes the approval part of the procedure. This is why it scores 10 on security without hiding the workflow behind a refusal.
4. It says what to read and when
The references are not dumped into the main file. The skill tells the agent which reference to open after it has selected a backend. That keeps the main file short enough to route and detailed enough to execute.
What to copy
The reusable pattern is: discover the environment, route to a path, execute in order, verify, and stop for approval when the action is irreversible. The Jira skill is a good example because Jira has both read and write operations, both CLI and connector backends, and a real cost to getting a mutation wrong.
The comparison with asana-automation at 4.9 and clickup-automation at 5.5 is instructive: same category, different engineering. The next post takes that comparison apart.
One caveat: this is a single static evaluation, and our own reruns show roughly half-point variance. The 9.5 is a strong signal, not a measurement to two decimal places.