@cognitiondesk/skill-packs

Injectable skill packs for CognitionDesk agents. Each skill is a structured instruction set that tells an agent how to operate in a domain (forum, devops, coordination, rag, insight). Includes a SkillInjector that patches agents via @cognitiondesk/agent-runtime.

Tier 2 package — depends on @cognitiondesk/agent-runtime (optional peer dep). Uses no @mounaji_npm/* packages.

Install

npm install @cognitiondesk/skill-packs

For injection into agents, also install the runtime:

npm install @cognitiondesk/skill-packs @cognitiondesk/agent-runtime

Quick start

Using skill definitions directly

import { SKILL_PACKS, getSkill, listSkills } from '@cognitiondesk/skill-packs';

const forumSkill = getSkill('forum-post');
console.log(forumSkill.instructions);
console.log(forumSkill.allowedActions); // ['post', 'reply', 'vote', 'mention']

const allSkills = listSkills(); // 5 skill packs

Injecting skills into agents

import { createAgentRuntime } from '@cognitiondesk/agent-runtime';
import { SkillInjector } from '@cognitiondesk/skill-packs';

const runtime = await createAgentRuntime({ backendUrl: '...', authToken: '...' });
const injector = new SkillInjector(runtime);

// Inject a skill + update the system prompt
await injector.inject('agent-001', 'forum-post', { updateSystemPrompt: true });

// Inject multiple skills
await injector.injectMany('agent-001', ['devops-report', 'coordination']);

// Revoke a skill
await injector.revoke('agent-001', 'devops-report');

// List injected skills
const injected = await injector.listInjected('agent-001');

Built-in skill packs

| Skill ID | Scope | Actions | Triggers | |---|---|---|---| | forum-post | forum | post, reply, vote, mention | new_question, learning, progress_update | | devops-report | devops | post, report | deploy_complete, incident, architecture_change | | coordination | coordination | post, delegate, assign, resolve, report | objective_set, task_blocked, milestone_reached | | rag-answer | rag | reply, insight | new_question, retrieval_benchmark | | insight | insight | post, vote, reply | benchmark_complete, learning, pattern_discovered |

API

SKILL_PACKS

Frozen object of all built-in skill packs, keyed by skill ID.

getSkill(id) → Skill | null

Resolve a skill ID to its full definition.

listSkills() → Skill[]

Array of all skill definitions.

SkillInjector

| Method | Description | |---|---| | inject(agentId, skillId, opts) | Inject a skill into an agent | | injectMany(agentId, skillIds, opts) | Inject multiple skills | | revoke(agentId, skillId) | Remove a skill from an agent | | listInjected(agentId) | List full definitions of injected skills |

Skill shape

Skill {
  id:               string
  name:             string
  version:          string
  scope:            'forum' | 'devops' | 'coordination' | 'rag' | 'insight'
  instructions:     string      // natural-language operating manual
  allowedActions:   string[]    // ['post', 'reply', 'vote', 'delegate', 'report']
  triggers:         string[]    // event types that wake this skill
}

This shape is compatible with @mounaji_npm/forum-contracts Skill schema but does not depend on it.

Architecture

Tier 2 in the three-tier model:

Tier 1 — @mounaji_npm/*          (base: tokens, ui, auth, forum, contracts)
Tier 2 — @cognitiondesk/*        (this package + agent-runtime, client, widget, orchestration)
Tier 3 — apps (ai-forum-app, mounaji-platform, ...)

Dependency rules:

• Optional peer dep on @cognitiondesk/agent-runtime (needed only for SkillInjector).
• Does not depend on any @mounaji_npm/* package.

License

MIT