Skill: Context Management

Purpose

Maintain and share project context across agent interactions, ensuring continuity and informed decision-making.

When to Use

• Orchestrator starts a new work session
• Agent needs project background for a task
• Handoff requires context transfer
• Resuming work after interruption

Components

Context management involves:

1. Project State - Current sprint, backlog status, recent activity
2. Decision History - Past decisions that inform current work
3. Active Work - In-progress items and their status
4. Blockers - Current impediments and dependencies

---

Context Sources

Primary Sources

Source	Information	How to Access
project.json	Project config, IDs	Read local file
Jira	Current sprint, backlog	Query via JQL
Confluence	Decision log, specs	Search via CQL
.env	User settings	Read local file

Context Refresh Triggers

• Start of new conversation
• Before major decisions
• When switching work contexts
• When agent reports stale information

---

Context Loading Protocol

1. Load Project Configuration

// Read project config
const config = require('./config/project.json');

// Extract key identifiers
const context = {
  project: config.project.name,
  jiraProject: config.jira.development.projectKey,
  confluenceSpace: config.confluence.spaceKey,
  cloudId: config.jira.cloudId
};

2. Load Current Sprint State

Atlassian:searchJiraIssuesUsingJql({
  cloudId: "{{jira.cloudId}}",
  jql: "project = {{jira.development.projectKey}} AND sprint IN openSprints()",
  fields: ["summary", "status", "assignee", "priority"]
})

3. Load Recent Decisions

Atlassian:searchConfluenceUsingCql({
  cloudId: "{{confluence.cloudId}}",
  cql: "space = '{{confluence.spaceKey}}' AND title ~ 'Decision Log' AND lastmodified > now('-7d')"
})

4. Load Blockers

Atlassian:searchJiraIssuesUsingJql({
  cloudId: "{{jira.cloudId}}",
  jql: "project = {{jira.development.projectKey}} AND labels = 'blocked' AND status != Done",
  fields: ["summary", "labels", "comment"]
})

---

Context Summary Format

When sharing context with agents or stakeholder:

## Project Context

**Project:** {{project.name}}
**Sprint:** Sprint {{number}} (Day {{day}} of {{total}})

### Sprint Status
- **Committed:** {{committed}} points
- **Completed:** {{completed}} points ({{percentage}}%)
- **In Progress:** {{inProgress}} items
- **Blocked:** {{blocked}} items

### Recent Decisions
{{#each recentDecisions}}
- **{{date}}:** {{summary}}
{{/each}}

### Active Blockers
{{#each blockers}}
- **{{key}}:** {{summary}} (blocked {{days}} days)
{{/each}}

### Key Contacts
- **Stakeholder:** {{STAKEHOLDER_NAME}}
- **Account ID:** {{ATLASSIAN_ACCOUNT_ID}}

---

Context Injection for Agents

When activating an agent, inject context:

## Agent Activation Context

You are the **{{agent.name}}** for {{project.name}}.

### Current Sprint
- Sprint {{sprint.number}}: {{sprint.goal}}
- {{sprint.daysRemaining}} days remaining
- Team velocity: {{sprint.velocity}} points

### Your Queue
{{#each agentQueue}}
- {{key}}: {{summary}} ({{status}})
{{/each}}

### Recent Activity
{{#each recentActivity}}
- {{timestamp}}: {{action}}
{{/each}}

### Relevant Decisions
{{#each relevantDecisions}}
- {{decision}}
{{/each}}

---

Context Persistence

Session Context

Maintain during conversation:

• Current topic/feature being discussed
• Decisions made this session
• Pending questions
• Files/tickets referenced

Cross-Session Context

Store in Jira/Confluence:

• All decisions → Decision Log
• All actions → Jira comments
• Specifications → Confluence pages
• Status changes → Jira transitions

---

Context Queries

Get Full Project Context

// Comprehensive context query
async function getProjectContext() {
  const [sprint, backlog, decisions, blockers] = await Promise.all([
    // Current sprint
    Atlassian:searchJiraIssuesUsingJql({
      cloudId, jql: "sprint IN openSprints()"
    }),
    // Backlog health
    Atlassian:searchJiraIssuesUsingJql({
      cloudId, jql: "sprint IS EMPTY AND status = Backlog"
    }),
    // Recent decisions
    Atlassian:getConfluencePage({
      cloudId, pageId: decisionLogId
    }),
    // Current blockers
    Atlassian:searchJiraIssuesUsingJql({
      cloudId, jql: "labels = blocked"
    })
  ]);

  return { sprint, backlog, decisions, blockers };
}

Get Feature Context

// Context for specific feature/epic
Atlassian:searchJiraIssuesUsingJql({
  cloudId: "{{jira.cloudId}}",
  jql: "project = {{projectKey}} AND 'Epic Link' = {{epicKey}}",
  fields: ["summary", "status", "description", "comment"]
})

---

Used By Agents

Agent	Context Usage
Orchestrator	Full context for routing and coordination
All Agents	Task-specific context for their work
Project Manager	Sprint and timeline context
Engineering Manager	Historical metrics and trends

Best Practices

1. Refresh before major decisions - Don't rely on stale context
2. Log context changes - When decisions change context, update logs
3. Minimal context transfer - Only pass what's needed for the task
4. Source of truth - Jira/Confluence are authoritative, not memory
5. Context validation - Verify assumptions before acting