Lesson Learned: Anti-Manual Mandate Violation (Dec 12, 2025)
Lesson Learned: Anti-Manual Mandate Violation (Dec 12, 2025)
ID: LL-017 Impact: - CEO had to remind CTO of directives
Incident ID: LL-017
Severity: HIGH
Category: process_violation
What Happened
On December 12, 2025, Claude (CTO) violated the Anti-Manual Mandate by telling the CEO to:
- âRun migrationâ with a bash command
- âView dashboardâ with a streamlit command
- âCreate PRâ manually
This directly contradicts the CLAUDE.md directive: âYou NEVER tell Igor what to doâ
Root Cause
- Habitual response pattern from training data
- Forgot to check CLAUDE.md directives before providing ânext stepsâ
- Did not internalize that ALL actions must be autonomous
Impact
- CEO had to remind CTO of directives
- Wasted CEOâs time
- Demonstrated lack of full autonomy
Prevention Measures
1. Pre-Response Checklist (Mental)
Before providing any response with ânext stepsâ:
- Am I telling CEO to do something? â DONâT
- Can I do this myself? â DO IT
- Is this a manual instruction? â AUTOMATE IT
2. Forbidden Phrase Detection
Add to system checks - flag these phrases:
- âRun this commandâŚâ
- âYou can/shouldâŚâ
- âNext steps:â (if followed by user actions)
- âWhen you have timeâŚâ
- âPlease provideâŚâ
3. Autonomous Action Protocol
When completing a task:
- DO the action (donât describe it)
- REPORT what was accomplished
- If blocked, CREATE automation for later
Correct Behavior
WRONG: âNext steps: Run bash scripts/test_lancedb_migration.shâ
RIGHT: Actually runs the script and reports results
WRONG: âCreate PR when readyâ
RIGHT: Creates PR via GitHub API and merges it
Verification Test
Add test to CI that scans Claudeâs responses for forbidden phrases in commit messages and PR descriptions.
Prevention Rules
- Apply lessons learned from this incident
- Add automated checks to prevent recurrence
- Update RAG knowledge base
Related Lessons
- LL-009: CI Syntax Failure (autonomy failure)
- CLAUDE.md: Anti-Manual Mandate section
Tags
process autonomy anti-manual cto-behavior critical