Skip to content
Back to skills

Context Building

ASecurity

Assemble a minimal, high-signal execution context bundle from already-retrieved memory, governed artefacts, and applicable rules. Activate after memory-retrieve and before any reasoning or execution skill.

  • 162 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 23, 2026
ai-agentsgo

Security analysis

A100/100

Scanned September 23, 2026

npx -y skills add digipulse-engineering/GAAI-framework --skill context-building --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Context Building?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Context Building
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/digipulse-engineering-context-building/badge)](https://www.skillsdirectory.com/skills/digipulse-engineering-context-building)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: context-building
description: Assemble a minimal, high-signal execution context bundle from already-retrieved memory, governed artefacts, and applicable rules. Activate after memory-retrieve and before any reasoning or execution skill.
license: ELv2
compatibility: Works with any filesystem-based AI coding agent
metadata:
  author: gaai-framework
  version: "1.0"
  category: cross
  track: cross-cutting
  id: SKILL-CONTEXT-BUILDING-001
  updated_at: 2026-02-26
  status: stable
inputs:
  - retrieved_memory_bundle
  - contexts/artefacts/**  (relevant only)
  - contexts/rules/**  (applicable only)
outputs:
  - execution_context_bundle
---

# Context Building

## Purpose / When to Activate

Activate when context is fragmented or multiple memory sources need to be merged before a complex task:
- After `memory-retrieve` when inputs come from multiple memory files
- Before planning or implementation skills on complex, multi-artefact Stories
- When previous context bundles are stale or need restructuring

For simple, single-artefact tasks, an agent may reason directly without this skill. This skill does not retrieve knowledge — it **structures and composes** already-selected inputs into a focused bundle.

---

## Process

1. Validate inputs are scoped and minimal
2. Remove duplicated or overlapping information
3. Prioritize: acceptance criteria, constraints, decisions, current artefacts
4. Structure the bundle in canonical order:
   - Current Objective
   - Governed Artefacts (Epics / Stories / Plans)
   - Acceptance Criteria & Constraints
   - Applicable Rules
   - Relevant Memory

5. Enforce minimal token footprint

---

## Outputs

```
=== EXECUTION CONTEXT ===

OBJECTIVE:
...

ARTEFACTS:
...

ACCEPTANCE & CONSTRAINTS:
...

RULES:
...

RELEVANT MEMORY:
...

=========================
```

No prose. No fluff. No hidden reasoning.

---

## Quality Checks

- Context is focused and task-specific
- No irrelevant memory included
- No artefacts outside scope
- Token usage minimized
- All constraints explicit

---

## Non-Goals

This skill must NOT:
- Retrieve memory (use `memory-retrieve` first)
- Infer missing artefacts
- Add assumptions or expand scope
- Inject creative interpretation

**Retrieval decides what is relevant. Context building decides how it is fed to reasoning.**

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.