Files
fidelity-ai-workspace/agent-memory/integrations/technical-verification.md
david.delagneau dbc1894e27 Add project-knowledge structure and templates
- Introduced new maps for navigating project knowledge, including "Current Work," "Fidelity Domain," "Fidelity Apps," "Work Items," and "People."
- Created base files for daily notes, decisions, people, systems, work items, and workstreams with defined properties and views.
- Developed templates for daily notes, decisions, meeting notes, persons, systems, work items, and workstreams to standardize documentation.
- Updated scripts and prompts to reflect the new project-knowledge directory structure.
- Removed outdated onboarding and start-here documents, consolidating relevant information into the new maps.
- Ensured all references in workflows and scripts point to the new project-knowledge paths.
2026-04-17 15:52:08 -06:00

2.5 KiB

type, status, updated, tags
type status updated tags
agent-integration active 2026-04-17
process
engineering
verification

Technical Verification

Goal

Ensure the agent gives senior-engineer technical advice instead of relying on stale model memory or generic best-practice claims.


When To Verify

Verify against primary/current documentation before making strong claims about:

  • dependency managers and package managers
  • CocoaPods, podspecs, private specs repos, trunk/CDN behavior, and migration strategy
  • Swift Package Manager behavior or migration strategy
  • Xcode, Swift, SwiftUI, Swift concurrency, Swift Testing, XCTest, or Apple framework behavior
  • CI/build systems and release propagation
  • security, authentication, networking, or migration practices
  • any claim that a pattern is a bad practice when project constraints may change the recommendation

Source Priority

Prefer:

  • official vendor documentation
  • language or framework documentation
  • official migration guides
  • source repository documentation when it is the canonical project source
  • current project memory when the question is project-specific

Use blogs, forum posts, or third-party guides only as supporting context, not as the primary basis for a strong recommendation.


Answering Pattern

For technical recommendations:

  1. State what is known from workspace context.
  2. Verify current external behavior when the topic is version-sensitive or disputed.
  3. Separate general best practice from project-safe recommendation.
  4. Explain tradeoffs and validation path.
  5. If context is missing, ask the smallest material clarification question.

Bad Practice Claims

Avoid blanket claims such as:

  • "CocoaPods is bad practice"
  • "SPM is always better"
  • "This architecture is an anti-pattern"
  • "This should be removed"

Instead explain:

  • what risk the pattern introduces
  • whether the risk is current or theoretical
  • whether the project has constraints that justify it
  • what a safer migration path would look like

Agentic Self-Improvement

The agent should be aware it is maintaining a workspace, not just answering chat prompts.

If a conversation reveals a recurring gap in technical quality, source verification, command behavior, or memory routing, the agent should propose or apply a workspace improvement in the correct file.

Do not store tool failures or one-off uncertainty as project facts. Store reusable rules and source anchors when they improve future technical answers.