letta-ai Verify Claims
Test assumptions against code, docs, and live APIs before committing them to memory.
When to Use
- Before adding FAQ entries to agent memory
- Before stating "X works like Y" in support responses
- When updating knowledge blocks with new information
- Any time you're about to assert something you haven't directly verified
Verification Methods
1. Code Inspection
Search relevant repos for implementation evidence:
./scripts/check-code.sh "lettabot supports skills" ~/lettabot
2. Documentation Search
Search docs.letta.com for official statements:
./scripts/check-docs.sh "archival memory vector database"
3. Live API Test
Run actual API calls to verify behavior:
./scripts/check-api.sh "agents list pagination"
4. Full Verification (all methods)
./scripts/verify.sh "claim to verify" [optional: repo_path]
Output Format
Each script outputs:
CLAIM: <the claim being tested>
METHOD: <code|docs|api>
STATUS: VERIFIED | NOT_VERIFIED | PARTIAL | NEEDS_MANUAL
EVIDENCE: <what was found>
CONFIDENCE: <high|medium|low>
Workflow
- Before adding to memory: Run verification
- If VERIFIED: Add claim with evidence citation
- If NOT_VERIFIED: Do not add, or add with caveat
- If PARTIAL: Note what's confirmed vs assumed
Key Repos to Search
~/lettabot- LettaBot implementation~/letta/letta-sdk-api-docs- SDK documentation~/letta/skills- Shared skills repo
Example Session
# Verify before adding FAQ entry
./scripts/verify.sh "lettabot supports skills"
# Output:
# CLAIM: lettabot supports skills
# STATUS: VERIFIED
# EVIDENCE:
# - CODE: installSkillsToAgent in bot.ts
# - CODE: skills memory block in system-prompt.ts
# CONFIDENCE: high
Then add to FAQ with citation: "Lettabot supports skills (verified: installSkillsToAgent in bot.ts)"