Matt Pocock skills
Matt Pocock skills explained: engineering practices, not an autonomous platform
Understand what mattpocock/skills actually distributes, why its 25 promoted skills differ from its full repository, and where the agent host still controls execution.
What you will learn
- Start with the deliverable
- Match a practice to a failure mode
- Evaluate a narrow, observable outcome
Before you start
- Basic repository, issue-tracker and test concepts
- An understanding that instructions and permission are different
Choose an adoption model and trace its files, authority boundaries and verification evidence.
Key takeaways
- The agent host executes work; the repository supplies practices and metadata.
- The promoted plugin set is smaller than the complete source tree.
- Judge a practice through a real acceptance condition, not output volume.
Start with the deliverable
mattpocock/skills packages written engineering and productivity practices for coding agents. The inspected revision contains Markdown skill instructions, supporting references, host metadata and a small set of maintenance scripts. It does not supply a model, application server, database or scheduler. A skill influences an existing agent; that agent and its tools still perform the work.
At commit 3cca18b, the plugin manifest and package metadata both report version 1.2.3. Its explicit list contains 25 promoted skills: 18 engineering and seven productivity entries. Eight in-progress and four miscellaneous skills also exist in the source tree. Repository presence therefore does not establish inclusion in the distributed plugin.
Match a practice to a failure mode
The useful unit is a repeatable decision or feedback practice: clarify an ambiguous request, establish domain terminology, define observable behavior, inspect a diff or preserve a handoff. For example, a shared definition of an order reservation can make a specification and its test names agree. Installing more instructions cannot repair a definition the team has never settled.
User-invoked skills organize deliberate sessions, while model-invoked skills supply reusable disciplines. This distinction explains why a router can help a person choose a flow without silently launching every flow it lists. It also keeps a short request from becoming permission for issue creation, commits or unrelated repository changes.
Evaluate a narrow, observable outcome
Try the collection on one bounded change in a disposable project and preserve the starting request, resulting specification, tests and review notes. Ask whether ambiguities were resolved and whether the implementation meets the stated behavior. These artifacts are stronger evidence than a fluent conversation or a large quantity of generated code.
The repository uses the MIT license. This review describes the pinned files and tests one maintenance script in isolation; it does not claim successful installation across every agent or a measured improvement in software quality. The remaining chapters distinguish documented workflows, observed script behavior and editorial proposals.
Implementation steps
- 1
Choose one failure mode from a recent small change.
- 2
Read the corresponding promoted skill and its references.
- 3
Use a disposable example with a written acceptance condition.
- 4
Compare the resulting specification, tests and review against that condition.
Copy-ready example
{
"sourceRevision": "3cca18b",
"pluginVersion": "1.2.3",
"promotedSkills": 25,
"inProgressSkills": 8,
"miscellaneousSkills": 4,
"agentInstallationVerified": false
}Frequently asked questions
Is this an agent runtime?
No. It distributes instructions and integration metadata for an existing agent host.
Are all 37 source-tree skills shipped in the plugin?
No. The inspected plugin explicitly lists the 25 engineering and productivity skills.
Sources
- README.mdSource checked 2026-09-08
- LICENSESource checked 2026-09-08
- package.jsonSource checked 2026-09-08
- .claude-plugin/plugin.jsonSource checked 2026-09-08
- .agents/invocation.mdSource checked 2026-09-08