Read Skill traces
Some Gateway deployments expose the Skill extension tool whose source name is
skill_traces. It combines the Skill trace-list, detail, spans, and statistics
reads behind an op discriminator. Use the exact namespaced name returned by
authenticated tools/list.
This compatibility tool is not declared by the current Agent Gateway main
branch. Call it only when it appears in your connected deployment’s
tools/list result. Current deployments can use the public Skill trace HTTP
endpoints instead.
- Required input:
skill_id in skill_... form.
- Operation: optional
op, one of list, get, spans, or stats;
defaults to list.
- Version scope: optional
version_ordinal; omission covers all versions.
- Trace scope:
trace_id is required for get and spans.
- List options:
start_time_unix_seconds, end_time_unix_seconds, limit,
cursor, sort, and user_id.
- Span options:
fields, limit, and cursor.
- Statistics:
request contains the statistics request body. Gateway owns the
Skill scope, so do not add a Skill fingerprint filter.
- Common Gateway fields can include
rationale and session_id.
- Effect: read only.
An organization administrator or the Skill’s creator can read permitted
cross-user results. Other members receive only their own traces. The Gateway
can return sliced cross-user traces limited to the Skill’s subtree.
