Agentweaver

Appearance

Project skills ​

Use project skills when a team member needs specialized reusable instructions, checklists, templates, or domain playbooks. Skills are added to the project catalog first, then assigned to individual agents.

Open the skills page ​

From a project, go to Skills or open /projects/:projectId/skills. The page has two tabs:

  • Catalog: add, inspect, and delete skills.
  • Assignments: choose which agents receive each skill.

Skills Catalog tab with catalog toolbar and skill cards

📸 Screenshot — skills-catalog.pngShows: the Skills page Catalog tab with Add Skill, Generate Skill, Import Skill, Sync connected repo, catalog status/provenance badges, assigned-agent chips, and View / Delete actions. Path: open a project → click Skills → /projects/:projectId/skills.

Add skills ​

The Catalog toolbar supports three acquisition paths:

  1. Add Skill: write a command slug, description, and instructions. Agentweaver saves a standard SKILL.md with name and description frontmatter.
  2. Generate Skill: describe the skill. The server generates a draft, then you review and edit it before creating the catalog skill.
  3. Import Skill: upload .md skill files, upload a folder with SKILL.md, or paste one of the supported source URLs — owner/repo, a https://github.com repo URL, a GitHub tree/blob folder URL, or a raw https://raw.githubusercontent.com/.../SKILL.md URL. Preview lists every candidate skill discovered at the source so you can select which ones to import.
  4. Browse marketplaces: select a trusted administrator-curated marketplace, search its available skills, then import selected candidates. Marketplace imports retain both the marketplace name and upstream GitHub repository/path in provenance.
  5. Sync connected repo: scans .github/skills, .copilot/skills, .claude/skills, and .agents/skills in the project repository.

Imports accept a single SKILL.md, a folder of <name>/SKILL.md directories, or recognized repo folders under .github/skills, .copilot/skills, .claude/skills, and .agents/skills.

Only GitHub sources are accepted

For safety against server-side request forgery, imports are restricted to two hosts: github.com (public HTTPS, default port, no embedded credentials) and raw.githubusercontent.com. Any other host — including http://, git@ SSH URLs, non-default ports, or user@host tricks — is rejected before anything is cloned or fetched. Only import skills from sources you trust, because imported instructions can change how an agent behaves.

Each catalog card shows status, provenance, updated time, source location, assigned agents, and a View action for the full instructions. Repeating an acquisition is safe: unchanged content is reported as unchanged, changed content updates the existing skill, and invalid content is rejected with validation errors.

Import Skill dialog with dropzones, trusted-source warning, URL field, and candidate preview controls

📸 Screenshot — skill-import-dialog.pngShows: the Import Skill dialog with the trusted-source warning, .md file and skill-folder dropzones, GitHub/raw URL field, Preview candidates, candidate selection, and Import. Path: /projects/:projectId/skills → click Import Skill.

Assign skills to agents ​

Open the Assignments tab after the project has a cast team. Each skill row shows a checkbox per agent. Check an agent to assign the skill; clear it to unassign. Each agent is labelled with its role alongside its name (for example, Kai — Backend Engineer) so you can tell cast members apart by what they do rather than only by name.

Only assigned and active skills appear for that agent at run time. The prompt shows the skill name, description, and path to the materialized SKILL.md; the full skill body is available lazily in the run worktree only if the agent needs it.

You can also drop skill files or a folder directly onto the Import dropzone. Single files come through the browser's normal file list, but folders are read through the FileSystem entry API and recursed so every <name>/SKILL.md inside is discovered with its correct relative path — dropping a whole skills/ directory imports each skill it contains. Binary and oversized (>1 MiB) files are skipped automatically.

See assigned skills on an agent ​

Assigned skills also appear on the agent itself. Open an agent from the Team page and its detail panel lists its Assigned skills — each skill's name and description — so you can review what an agent is equipped with without switching to the Skills page. If nothing is assigned, the panel says No skills assigned.

See also ​