read-memorylisted

# read-memory Loads the topic's memory sections so the agent can answer with cited facts. ## Behavior Run the dump script — **one Bash call, do not read the files individually**: ``` bash .claude/skills/read-memory/dump.sh ``` It prints all six sections (`background`, `current_understanding`, `decisions`, `open_questions`, `commitments`, `people`) under `## <section>` headers, each with its last-commit timestamp, and `(尚未有任何條目)` for empty sections. ## When to invoke - Every Q&A turn (auto-load at start of conversation) - When user asks "what do we know about X" - Before writing memory (to avoid duplicating existing entries) - Before answering — never speak from training data alone ## What NOT to do - Don't read memory/*.md one file at a time — that's 12 tool calls for what the script does in one - Don't paraphrase — use the raw markdown so citations stay verbatim - Don't filter by relevance — let the agent decide what to use - Don't fetch sources/ — that's for `ingest-source`, not for normal Q&A