Skills
SOBA bundled, user, and project skills with eval, bench, trace, promotion, history, and rollback.
Skills are Markdown instruction packages for repeatable agent behavior. In 0.6.x, they are not just prompt snippets: they have a local lifecycle with evals, bench summaries, traces, revision history, promotion, and rollback.
Skill scopes
| Scope | Location | Use |
|---|---|---|
| Bundled | skills/ in the SOBA package | Always available |
| User | ~/.soba/skills/ | Available across projects |
| Project | .soba/skills/ | Available only in the current project |
| Internal | .soba/skills/ inside the SOBA repo | Used while developing SOBA itself |
When names collide, SOBA resolves them as: Project > User > Internal > Bundled.
Commands
/skill list
/skill new <name>
/skill edit <name>
/skill eval <name>
/skill bench <name>
/skill trace <name>
/skill promote <name>
/skill history <name>
/skill rollback <name> <revision-id>
/skill rm <name> --confirm
/skill:<name> [prompt]eval runs deterministic skill evaluation cases. bench shows aggregate quality: run count, success rate, median pass
rate, latest run, and common failures. trace opens recent eval runs and failed cases.
These commands make a skill closer to a small behavior package than a loose custom instruction.
Skill eval bench and trace
Use this loop when a skill is meant to be reused:
- Run
/skill eval <name>after creating or editing the skill. It executes deterministic evaluation cases and stores local eval artifacts. - Run
/skill bench <name>to inspect aggregate quality: run count, success rate, median pass rate, latest run, and common failures. - Run
/skill trace <name>when a bench result is unclear. Trace shows recent eval runs and failed cases, so the next edit can target the actual failure.
Do not promote a project skill until eval, bench, and trace tell the same story.
Skill card checklist
For production-quality skills, keep these sections explicit:
- purpose and trigger examples;
- false-positive examples;
- required tools and forbidden actions;
- verification contract or required checks;
- known failure modes;
- eval tasks.
Example:
---
name: fix-until-green
description: Diagnose a failing project check, fix the root cause, and rerun the targeted check before broader gates.
---
## Trigger Examples
- "Fix this failing test output."
- "Run the targeted check and make it green."
## Verification Contract
- Reproduce the failure first.
- Prefer a targeted test before a full suite.
- Do not call the task complete while the target check fails.