Cross-project architectural knowledge system for developers
Context Engine has 14 well-named tools with clear descriptions and explicit Zod schemas. Naming follows verb_noun convention (store_concept, search_concepts, get_concept, etc.). Descriptions are detailed and contextual (avg ~150 chars), explaining WHEN to use each tool. However, output schemas are not documented, responses are inferred from handler implementations, not visible in the tool registration. Parameter descriptions are present but some lack format constraints (e.g., 'limit' has no min/max). Error handling is minimal, no recovery guidance or actionable error messages visible. Security is reasonable (no secrets in params), but no permission gates or audit logging. Composition is strong: tools are single-responsibility and chainable (e.g., search_concepts → get_concept → link_concepts). Overall: solid definitions with good naming and descriptions, but incomplete schema documentation and weak error handling prevent a higher score.
Remove an outdated or incorrect concept.
Discover how a project connects to others through shared concepts. Useful for understanding dependencies between projects in your ecosystem.
Retrieve a specific concept with all its linked concepts and full context.
Analyze what projects would be affected by changes to a concept. Use this before making significant changes to understand the blast radius.
Get full context about a project: type, framework, detected patterns, key files, and related concepts. Use this to understand a project before starting work.
Get the filesystem path for a project. Use this path with Glob/Grep/Read tools to search the codebase.
Output schemas not documented. Tool handlers return structured data, but response field names, types, and pagination are not declared in tool registration. LLMs cannot plan downstream calls or extract fields reliably.
Numeric parameters lack constraints. 'limit' in search_concepts and list_concepts has no min/max documented. LLMs may pass absurd values (0, 999999) that break pagination or exhaust resources.
Error handling is silent. No visible error recovery guidance, categorization (retryable vs fatal), or actionable error messages. If a concept is not found or a link fails, LLMs have no guidance on next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Find concepts that are shared between two projects - useful for understanding how projects connect.
Get an overview of all projects and concepts in context-engine. Shows a table with each project's type, detected patterns, and concept count. Use this to get a quick understanding of what's indexed.
Create a relationship between two concepts with context explaining why they're linked. Relationship types: - uses: Source uses functionality from target - requires: Source depends on target - extends: Source builds upon target - contradicts: Source conflicts with target - related: General relationship
List all stored concepts, optionally filtered by project. Check what's known about a project before starting work.
List all indexed projects with their detected patterns and frameworks.
Search for relevant architectural knowledge before making changes. Use this to check for related knowledge from previous sessions: - Before modifying authentication - Before changing database schemas - Before updating API contracts - When exploring how systems connect
Store architectural knowledge about cross-project patterns, dependencies, or decisions. Call this when you discover insights worth preserving for future sessions: - Cross-project dependencies found during debugging - Authentication/authorization flows - Data flow patterns between services - Migration decisions and rationale - Integration points and shared contracts - "Gotchas" and edge cases discovered during work
Append new learnings to an existing concept or update its content.
Destructive operations (delete_concept) lack confirmation or dry-run support. Agents can permanently delete concepts without a safety gate, risking data loss.
No permission gates or audit logging. Tools do not declare required scopes (read:concepts, write:concepts, delete:concepts) or log who called what. Compliance and incident response are blind.