Skill activation
The point when an agent judges a discovered skill relevant and loads its instructions into the current task.
- Published
- Sep 27, 2026
- Checked
- Sep 27, 2026
How far we checked this page
Signed by Ravalika Korthiwada, Sep 27, 2026. Passing one check says nothing about the next.
- FormatNot checkedThe file format was not checked for this page.
- InstallsNot checkedInstallation was not checked for this page.
- LoadsNot checkedLive loading was not checked for this page.
- HelpsNot checkedA with/without task comparison was not run.
Skill activation is the moment an agent decides a discovered agent skill applies to the current request and loads its instructions, rather than just knowing the skill exists. It’s a distinct step from installing a skill (placing its files on disk) and from discovering it (the host reading its frontmatter into its list of available skills): a skill can be installed and discovered without ever activating, if nothing the user asks matches its description.
We measured this directly on Codex CLI 0.157.1 on September 27, 2026, comparing our explain-error skill (a specific description naming errors, exceptions, stack traces and tracebacks) against a copy with a vague description, Helps with errors., and otherwise identical instructions. Across three matching prompts (Python, JavaScript and Java error traces), the specific description activated 3 of 3 times; the vague one activated only 1 of 3, missing the Python and Java prompts. Neither version produced a false activation on three unrelated prompts. The sample is small, three prompts per skill: enough to see a real difference, not enough to generalize an exact activation rate.
Related: Agent skill, Progressive disclosure, Frontmatter.