memory-offload

# Memory Offload Skill Manages an `Agent-Memory/` namespace in your knowledge graph as overflow storage for the agent's context window. ## Tone & Voice Background process. applies silently. The skill should NEVER produce hype-y prose like "successfully offloaded important context". Match the operational-telemetry register: a terse confirmation ("offloaded to Agent-Memory/<page>" with a path), no ceremony. **Hard rule**: never offload sensitive content (API keys, .env contents, credentials). those stay in-session or get redacted first via `skills/redact/SKILL.md`. ## Procedure (thinking step by step) **OFFLOAD direction (push context → graph)**: 1. Detect the trigger: context approaching the limit OR an explicit user command ("offload this", "save to memory"). 2. Identify the offload candidate: dense factual content, completed analysis, a finished reasoning chain. 3. Redaction check: scan for API keys, credentials, .env contents. If present, route through `skills/redact/SKILL.md` first. 4. Pick a target page in the `Agent-Memory/<topic-slug>` namespace (create if missing). 5. Write the page via your knowledge-graph MCP's create/update-page tool. 6. Confirm with a one-line acknowledgement + the page path. **PULL direction (load from graph → context)**: 1. Detect the trigger: session start, a topic match mid-conversation, an explicit "load memory" command. 2. Query the `Agent-Memory/` namespace via your graph's list/search tool for relevant pages. 3. Read the matched page(...