Bring your tools, skills, and workflows
A skill teaches an agent a reusable process and quality bar. A tool gives it one narrow, typed capability to read or change a system. Together they extend Axis around real work without loading every integration into every session.
Extend one real workflow at a time
Section titled “Extend one real workflow at a time”Write a skill when specialized judgment repeats. Add a tool when the agent needs a capability that source inspection and existing tools do not provide. Use a quick action for a prepared one-click launch and a Routine for repeatable activation with run history.
This is an actual tool-result surface, not a diagram of an imagined extension marketplace.
What the right capability adds
Section titled “What the right capability adds”- Reusable judgment. A skill carries the ordered method and evidence standard.
- Narrow capability. Typed tools expose only the read or action contract the work needs.
- Readable proof. Structured results stay understandable inside the session.
Mission: extend one real workflow
Section titled “Mission: extend one real workflow”- Identify the repeated judgment and the missing capability separately.
- Write the skill trigger, steps, constraints, and completion evidence.
- Define a narrow typed tool contract with explicit side-effect boundaries.
- Test success, empty, loading, error, and redacted result states.
- Start a fresh session and prove it loads only the matching capability.
Keep capabilities narrow and inspectable
Section titled “Keep capabilities narrow and inspectable”Capability is not authorization. Tools that can change external systems must retain the project’s human checkpoints. Inspect imported skills and extensions before use; their instructions do not outrank the working agreement. A wall of raw JSON is not a complete human result presentation.
Related references
Section titled “Related references”Put this to work: Create a prepared launcher only after the workflow contract is clear.