Qredence Documentation

Skills are versioned instructions and resources that guide Fleet through a specific kind of work. Fleet ships a fixed, trusted catalog of bundled Skills.

Bundled Skills

Skill	Version	Use it for
`data-analysis`	1.2.0	Compute and verify descriptive statistics, trends, and qualified anomalies.
`dspy-rlm`	1.1.0	Analyze, explain, or implement `dspy.RLM`. Not for RAG or `dspy.Retrieve`.
`long-context`	2.3.0	Discover public sources and analyze large documents, transcripts, code, or datasets with Sandbox Python.
`report-builder`	1.3.1	Create, save, read back, and verify reports from trusted source data.
`workspace-files`	1.4.1	Use the Session Workspace, Project, Attachment, and Artifact tools correctly.

How Skills load

Every Turn starts with short Skill Cards that advertise the catalog.
Without an explicit selection, Fleet may load up to four advertised Skills with load_skill.
After a Skill loads, read_skill_resource returns one of its declared resources.

When the Turn has Session Workspace access, loading a Skill also installs its resources at skills/<name>/<path> in the Session Workspace, so generated code can read or run them. Each Skill Card lists affordances, such as workspace.files or artifacts.publish. Affordances are guidance only. They don’t control which host tools exist.

Select Skills for a Turn

In the terminal, list the catalog and pin a Skill for the next Turn.

/skills
/skill report-builder

Clear the selection with /skill clear. Over HTTP, pass up to four exact, version-pinned selections in skill_selections.

{
  "text": "Build a summary report from the attached CSV.",
  "attachment_ids": ["<attachment-uuid>"],
  "skill_selections": [
    { "id": "<skill-uuid>", "expected_version": "1.3.1" }
  ]
}

An explicit selection preloads those Skills and restricts the Turn to that set. List Skill IDs with GET /api/skills.

Last modified on October 9, 2026