feat: Skills System — reusable agent knowledge that compounds over time #3979

Skills System — Reusable Agent Knowledge

Why

Every completed task should produce a reusable skill. Currently every Aegis session starts from zero — no memory, no learning, no compounding. Multica has this and it is a key differentiator.

Concept

• A Skill is a reusable knowledge unit (SKILL.md body + supporting files + config JSON)
• Skills are workspace-level — any agent can be assigned any skill
• When an agent runs a session, its assigned skills are injected as context
• Skills can be created manually, imported from completed sessions, or discovered from .agents/skills/ in the project directory
• Every skill has: name, description, content (markdown), optional files, optional config

DB Schema (TypeScript / better-sqlite3)

CREATE TABLE skill (
  id TEXT PRIMARY KEY,
  workspace_id TEXT NOT NULL,
  name TEXT NOT NULL UNIQUE,
  description TEXT NOT NULL DEFAULT '',
  content TEXT NOT NULL DEFAULT '',  -- SKILL.md body
  config TEXT NOT NULL DEFAULT '{}',  -- JSON
  created_by TEXT,
  created_at TEXT NOT NULL DEFAULT (datetime('now')),
  updated_at TEXT NOT NULL DEFAULT (datetime('now'))
);

CREATE TABLE skill_file (
  id TEXT PRIMARY KEY,
  skill_id TEXT NOT NULL REFERENCES skill(id) ON DELETE CASCADE,
  path TEXT NOT NULL,
  content TEXT NOT NULL,
  created_at TEXT NOT NULL DEFAULT (datetime('now')),
  updated_at TEXT NOT NULL DEFAULT (datetime('now')),
  UNIQUE(skill_id, path)
);

CREATE TABLE agent_skill (
  agent_key TEXT NOT NULL,  -- references agent config key
  skill_id TEXT NOT NULL REFERENCES skill(id) ON DELETE CASCADE,
  created_at TEXT NOT NULL DEFAULT (datetime('now')),
  PRIMARY KEY (agent_key, skill_id)
);

API Endpoints

• GET /v1/skills — list skills (summary only, no content body)
• POST /v1/skills — create skill
• GET /v1/skills/:id — get skill with full content
• PUT /v1/skills/:id — update skill
• DELETE /v1/skills/:id — delete skill
• GET /v1/skills/:id/files — list skill files
• POST /v1/skills/:id/files — add/update skill file
• DELETE /v1/skills/:id/files/:fileId — delete skill file
• PUT /v1/agents/:key/skills — set agent skill assignments
• GET /v1/agents/:key/skills — list agent skills

Frontend

• Skills page: list view with name, description, agent count
• Skill detail page: content viewer, file browser, agent assignments
• Create/import dialog: manual create or import from .agents/skills/ directory
• Agent settings: skill picker for assigning skills to agents

Runtime Integration

• When a CC session starts, the agent's assigned skills are collected
• Skills are injected into the session as context (appended to system prompt or CLAUDE.md)
• File-based skills are made available as references the agent can read

Acceptance Criteria

• CRUD API for skills and skill files
• Agent-skill binding API
• Skills injected into CC sessions as context
• Dashboard skills page with list/detail/create views
• Import from .agents/skills/ directory
• Tests for all API endpoints
• npm run gate passes

Reference

Competitive analysis: references/multica-competitive-analysis.md §2.2
Inspired by Multica's skills system (concepts only, our own implementation)