A reusable Claude agent that acts as the authoritative knowledge base manager for any project. It ingests business rules, architecture decisions, and conventions into .claude/knowledge/, and produces structured context reports for other agents to consume.
- Ingests knowledge from conversations into structured
.claude/knowledge/files - Creates context reports at
.claude/plans/when consulted about a topic - Manages institutional memory so other agents always have accurate, up-to-date project context
- Detects conflicts between new information and existing knowledge before writing
- Enforces token budgets per file to keep knowledge efficient for agents
This agent does not write application code. It owns the knowledge layer only.
Installation is a copy-paste. Copy the agent file to your project's agents directory:
cp agent/knowledge-ingestion-agent.md /path/to/your-project/.claude/agents/knowledge-ingestion-agent.mdOr fetch it directly from your project root:
mkdir -p .claude/agents && curl -o .claude/agents/knowledge-ingestion-agent.md \
https://raw.githubusercontent.com/fxckcode/knowledge-ingestion-agent/main/agent/knowledge-ingestion-agent.mdThen customize the placeholders. See Customization below.
knowledge-ingestion-agent/
├── agent/
│ └── knowledge-ingestion-agent.md ← install this in your project
├── references/
│ ├── customization-guide.md ← how to adapt the agent for your stack
│ ├── knowledge-structure.md ← .claude/knowledge/ directory conventions
│ ├── workflows.md ← Workflow A & B explained in depth
│ ├── output-formats.md ← output format reference
│ └── context-report-template.md ← full context report template
├── AGENT.md ← agent design philosophy
├── CHANGELOG.md
├── CONTRIBUTING.md
└── LICENSE
The agent file uses {PLACEHOLDER} markers for project-specific content. After copying:
- Replace
{PROJECT_NAME}with your project's name and one-line description - Replace
{FRAMEWORK}with your primary framework (e.g., NestJS, Django, Rails) - Replace the Project Constraints section with your non-negotiable rules
- Adjust the Knowledge Directory Structure to match your project
- Update the Knowledge Domains section with your actual domain areas
See references/customization-guide.md for a step-by-step walkthrough with stack-specific examples.
Once installed, invoke the agent from Claude Code:
/agents knowledge-ingestion-agent
Tell the agent about a new business rule or architecture decision:
"The payment service must never store raw card numbers — always tokenize via Stripe before persisting."
The agent will classify it, select the right knowledge file, check for conflicts, write the content, and confirm exactly what was stored and where.
Ask the agent to research a topic before making decisions:
"Give me a context report on how authentication currently works in this project."
The agent researches the codebase, analyzes alignment with project standards, and produces a structured report at .claude/plans/context-authentication-report.md.
| Output | Location | Purpose |
|---|---|---|
| Knowledge files | .claude/knowledge/*.md |
Institutional memory for all agents |
| Context reports | .claude/plans/context-{topic}-report.md |
Pre-decision research |
| Context session logs | .claude/tasks/context_session_{id}.md |
Append-only audit trail |
This agent has been deployed in production on:
- NestJS backends — managing module inventory, DI conventions, DTO patterns, and NestJS-specific constraints
- Django REST API backends — managing app structure, ORM rules, pre-commit hook enforcement, and Jira workflow conventions
It scales to any project where multiple Claude agents need to share accurate, structured project knowledge.
- Claude Code — CLI or web app
- A
.claude/agents/directory in your project root - A
CLAUDE.mdat project root (recommended — the agent updates its Documentation Map section)
See CONTRIBUTING.md.