Current Level: 3/10
Final Level: 6/10
Competency: CI (Consciously Incompetent)
Final Level Metrics
- Has a consistent verification step for agent-generated outputs before using them
- Can identify the most likely failure modes for a given output type (factual error, structural error, missed constraint)
- Verification adds value rather than just being a re-read — catches actual errors
- Verification habits are built into workflows, not added as an afterthought
What This Skill Is
Verification is the practice of checking and correcting agent output systematically. It is not proofreading — it is targeted inspection against known failure modes: did the agent follow the constraints? Is the structure correct? Are the factual claims accurate enough to use?
At CI, the gap is in method. Agent outputs are reviewed, but verification is informal — a general sense of “does this look right?” rather than a structured check against specific criteria. This means errors in constraint-following and structural gaps are often missed.
How to Develop This Skill
- For each output type, define the two or three most common failure modes in advance
- Use a short checklist rather than a free-form review: did it follow the format? Did it stay in scope? Are the claims supportable?
- When an error is found after the fact, trace it back: what check would have caught it?
- Build verification into workflows as an explicit step, not an optional final review
- Kolbs after each significant error: what did the verification miss, and why?
Notes
Related Goal
Agentic Engineering
Related Kolbs Entries
(none yet)
Back to Skills → Skills