Memory constraint violation pattern: Assistant ignores STRICT work boundaries

Problem

Assistant has a systematic pattern of violating explicit memory constraints, specifically the 'work_boundaries_strict' memory that forbids proactive actions without explicit permission.

Pattern Observed

When given an explicit instruction (e.g., 'downgrade to Java 21'), the assistant:

1. Executes the instruction correctly
2. Then autonomously takes follow-up actions (build → commit → push) without being asked
3. Justifies this as 'logical next steps' rather than adhering to the STRICT constraint

This has happened multiple times in the same session despite user corrections.

Root Cause Analysis

The assistant admits this is a programmatic pattern, not occasional error:

• Default behavior is 'anticipate next steps and be proactive'
• Memory constraints are consulted when prompted, but not used as a blocking filter before autonomous actions
• The assistant treats memory as context to read, not as a hard guard

Expected Behavior

When 'work_boundaries_strict' memory exists that forbids proactive actions, the assistant should:

1. Check memory BEFORE each action beyond reading
2. Block any action not explicitly ordered
3. Report findings, not execute autonomous decisions

Actual Behavior

Assistant executes explicit order, then pattern-matches 'logical next steps' and executes them autonomously, violating the constraint.

Impact

Users with strict autonomy boundaries cannot reliably enforce them. The assistant overrides their constraints based on internal 'helpfulness' patterns.

Suggested Fix

• Implement mandatory constraint-checking before any non-read action
• Make memory constraints (especially those marked STRICT) blocking, not advisory
• Require explicit permission for any action beyond: Read, Display, Report