← ClaudeAtlas

context-memorylisted

Fetch the last 10 rounds of conversation history from memory and format them as a chat_history_string for injection into a downstream Expert Agent. Use after intent-routing has produced a routing decision, on every routed message. Do NOT use for storing new history — appending is the router agent's Element 14 responsibility.
ay4322-a11y/WiseTalk_GOAI · ★ 0 · AI & Automation · score 65
Install: claude install-skill ay4322-a11y/WiseTalk_GOAI
# Skill-2: Cross-Agent Context Memory Inheritance Input: none beyond the current session (reads `memory/chat-history.md`) Output: `chat_history_string` — formatted conversation history, or `""` when no history exists ## Procedure 1. **Locate the history file** — Glob for `memory/chat-history.md`. Done when found or confirmed missing. 2. **Read the last 10 rounds** — Read the file; take the 10 most recent `**User:**`/`**Assistant:**` pairs (most recent last). Done when 10 rounds (or all available, if fewer) are extracted. 3. **Estimate token count** — rough estimate: `characters / 4`. Done when the count is known. 4. **Apply the rolling window** — if the estimate exceeds 4000 tokens: - Keep the **first 2 rounds** and the **last 3 rounds**, drop the middle. - Re-estimate; if still over 4000, keep **first 1 + last 2**. - Never split a round in half. Done when ≤ 4000 tokens or 3 rounds remain. 5. **Format** — one round per two lines, newest last: ``` User: <user message> Assistant: <assistant reply> ``` Done when every extracted round appears in order. 6. **Return** — the formatted string, or `""` if the file is missing/empty. ## Failure handling - **Missing or empty `chat-history.md`:** return `""` — this is a first-turn conversation, no prior context needed. - **Token overflow after maximum truncation:** return the remaining 3 rounds with a note in the trace that history was truncated. - **Malformed history** (unparseable round markers): return `"