Memory Gate

Memory Gate

The Memory-First Gate and its investigation protocol, extracted from the memory router per ADR-063 (memory-skill decomposition). memory still routes here; an agent about to change an existing system loads this sub-skill instead of the full memory surface.

The gate is one operation with a single rule: before you change something that already exists, search memory for why it exists. The search uses the canonical Tier 1 script; this sub-skill owns the enforcement semantics (ADR-070) and the Chesterton's Fence framing, not the search implementation.

Triggers

Use this skill when the user says:

• memory-first gate to apply the BLOCKING pre-change check
• search memory before changing to recover historical context first
• chesterton fence check to investigate why a system exists before removing it

Memory-First Gate (BLOCKING)

Before changing existing systems, you MUST:

1. Search memory for the topic you are about to change.
2. Review results for historical context.
3. If Tier 1 is insufficient, escalate to Tier 2 (episodes).
4. Document findings in your decision rationale.
5. Only then proceed with the change.

# Recover the "why" before you change the "what"
SCRIPTS_DIR="${COPILOT_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-.claude}}/skills/memory/scripts"
python3 "$SCRIPTS_DIR/search_memory.py" "[topic]"

Why BLOCKING: under 50% compliance with "check memory first" guidance when it is advisory. Making it BLOCKING achieves 100% compliance, the same pattern as the session protocol gates (ADR-070 gate semantics).

Verification: the transcript, pull request, handoff, Serena memory, or an optional log must show the search before the decision.

The canonical Tier 1 script is search_memory.py. It is shared with the memory router and lives at .claude/skills/memory/scripts/search_memory.py. This sub-skill delegates to it and does not reimplement search.

Memory-First as Chesterton's Fence

Core insight: memory-first architecture implements the Chesterton's Fence principle for AI agents.

"Do not remove a fence until you know why it was put up" (G.K. Chesterton)

Translation for agents: do not change code, architecture, or protocol until you search memory for why it exists.

Why this matters

Without memory search (removing the fence without investigation):

• Agent meets complex code, thinks "this is ugly, I will refactor it".
• Removes validation logic that guards an edge case.
• A production incident follows.
• Memory held the past incident that explained why the validation existed.

With memory search (Chesterton's Fence investigation):

• Agent meets complex code.
• Searches memory: search_memory.py "validation logic edge case".
• Finds the past incident that explains why the code exists.
• Makes an informed decision: preserve, modify, or replace with equal safety.

Investigation Protocol

When you want to change something that already exists, match the change type to the required search:

What memory contains (git archaeology)

Tier 1 (Semantic): facts, patterns, constraints.

• Why is Python the default for new internal automation? (ADR-042, which superseded the PowerShell-only ADR-005 but keeps PowerShell for quick fixes to existing PowerShell scripts and PowerShell-specific operations)
• Why did the PowerShell-only constraint exist, before it was superseded? (ADR-005, superseded by ADR-042)
• Why do skills exist instead of raw CLI? (usage-mandatory)
• What incidents led to BLOCKING gates? (protocol-blocking-gates)

Tier 2 (Episodic): past session outcomes.

• What happened when we tried approach X? (session replay)
• What edge cases did we hit? (failure episodes)

Memory is your investigation tool. It holds the "why" that Chesterton's Fence requires you to discover before you act. For the full four-phase decision framework (Investigation, Understanding, Evaluation, Action) and the decision matrix for when to investigate, use the chestertons-fence skill.

Escalation

Tier 1 answers most "why does this exist" questions. Escalate when it does not:

Change target identified?
│
├─► Tier 1: search_memory.py "[topic]"
│   └─► Facts, constraints, ADR rationale. Enough? Proceed.
│
└─► Not enough context? Escalate to Tier 2 (episodes)
    └─► What happened last time we touched this, and how did it end?

Escalate only when the cheaper tier is insufficient. Starting at Tier 1 keeps the token cost low; most gate checks resolve there.

Verification

Anti-Patterns

Process

Phase 1: Identify

Name the existing system you intend to change and the change type. Match it to a row in the Investigation Protocol table.

Phase 2: Search

Run search_memory.py for the topic. Escalate to Tier 2 only when Tier 1 is insufficient.

Phase 3: Decide

Record the findings in your decision rationale, then preserve, modify, or replace with equal safety. Never proceed with the change before the search is logged.

Related Skills

References

• ADR-007: Memory-first architecture (the posture this gate enforces)
• ADR-037: Memory router architecture (the router this sub-skill delegates from)
• ADR-063: Memory skill decomposition (this extraction)
• ADR-070: Gate semantics (why the gate is BLOCKING, not advisory)
• references/agent-integration.md: multi-agent integration patterns, including the Memory-First Decision Making workflow this gate enforces