T2A Memory

Instructions

Use this skill when the user asks you to install T2A Memory, remember a fact, or use user-level memory.

1. Run {{T2A_MEMORY_BIN}} remember --memory-home {{MEMORY_HOME}} --tool-home {{TOOL_HOME}} "<fact>".
2. Read the emitted t2a.handoff.
3. Use the generic t2a skill if it is available to resolve the capsule.
4. Fallback workflow: read task.json, prompt.md, input.json, and output.schema.json; write only output.json.
5. Before commit, run core T2A validation and T2A Memory validation:
   • {{T2A_BIN}} validate "<task-dir>"
   • {{T2A_MEMORY_BIN}} validate --memory-home {{MEMORY_HOME}} --tool-home {{TOOL_HOME}} "<task-dir>"
6. If both validations pass, run {{T2A_MEMORY_BIN}} commit --memory-home {{MEMORY_HOME}} --tool-home {{TOOL_HOME}} "<task-dir>".
7. After commit, verify with {{T2A_MEMORY_BIN}} status --memory-home {{MEMORY_HOME}} and {{T2A_MEMORY_BIN}} read --memory-home {{MEMORY_HOME}}.
8. Do not rerun {{T2A_MEMORY_BIN}} validate on the same already committed task as the normal close-out check.

Guidelines

• Do not edit memory.json or MEMORY.md directly.
• Do not use network access during memory handoff or consolidation.
• Do not call LLM APIs or install LLM SDKs for this workflow.
• Do not ask for model, base_url, api_key, provider settings, or vector database configuration.
• Do not run git commit unless the user separately asks for a git commit.
• t2a-memory commit commits to local memory files only. It is not git commit.

Examples

• "Install T2A Memory for yourself and remember that T2A tools should not embed LLM API keys."
• "Remember this preference in T2A Memory."
• "Use T2A Memory to store this durable local-memory fact."