Skills¶

Skills are packaged instructions an agent loads when it needs them: a folder with a SKILL.md and any templates or references it uses, the same idea as Claude Code skills.

---
name: research-report
description: Plan, evidence and write a rigorous research report with inline citations. Use for ...
allowed-tools: [web_search, web_fetch, notes_write]
---
# Research report
1. Frame ... 2. Gather ... 3. Analyse ... 4. Write using templates/report.md ...

Progressive disclosure¶

sequenceDiagram
    participant A as Agent
    participant R as Runtime
    R->>A: system prompt lists every permitted skill's name and description
    A->>R: use_skill("research-report")
    R-->>A: the full SKILL.md instructions
    A->>R: skill_read("research-report", "templates/report.md")
    R-->>A: the bundled file

Agents see only names and descriptions until they call use_skill, so a large library costs little context. skill_read opens files bundled with the skill.

Where skills come from¶

Source	Skills
Built-in	`research-report`, `latex-paper`, `code-review`, `python-project`, `source-evaluation`, `executive-summary`
Marketplace	`adr-writer`, `api-design-review`, `changelog-writer`, `data-analysis-report`, `incident-postmortem`, `literature-review`, `test-plan`, `threat-model`
Yours	Created in the editor, with `POST /api/v1/skills`, or imported from a zip

A skill of yours with the same name as a built-in overrides it for your runs. Built-in skills are read-only and cannot be deleted.

Marketplace¶

The marketplace is a curated catalogue that ships with Argus (backend/argus/skills/marketplace/). GET /api/v1/skills/marketplace lists it with each skill's version, tags, allowed tools and files, and whether you have installed it; GET /skills/marketplace/{name} shows the full SKILL.md and files. POST /skills/marketplace/{name}/install copies a skill into your library, after the same validation as an upload, and you can then edit it like any skill of your own.

Managing your skills¶

Endpoint	Purpose
`POST /skills/validate`	Check a `SKILL.md` and its files without saving
`POST /skills`	Create a skill
`POST /skills/import`	Import a zip
`PUT /skills/{name}`	Replace a skill (records a new version)
`PATCH /skills/{name}`	Enable or disable it
`DELETE /skills/{name}`	Delete it and its history
`GET /skills/{name}/export`	Download it as a zip
`GET /skills/{name}/usage`	Which councils use it (`used_by_councils`)
`GET /skills/{name}/versions`	Version history: number, semantic version, author, note, files
`GET /skills/{name}/versions/{n}`	One stored version
`POST /skills/{name}/versions/{n}/restore`	Restore a version (recorded as a new version)

Writes need the skills:write scope. Uploads are validated: frontmatter rules, text files only, no hidden files or path traversal, and size limits checked before a zip is extracted.

Which skills a council may use¶

Guardrails decide, with glob patterns:

{"skills": {"allow": ["*"], "deny": ["latex-*"]}}

A disabled skill is not offered to agents. Sub-councils inherit the parent's skill deny list.