propose-skilllisted

# Propose Skill > **Best with the `multiplai-context` plugin.** Steps below read that plugin's runtime artifacts (`.multiplai/diary/` and `.multiplai/learnings/`). Without it, rely on the current conversation context alone. Detect repeating patterns in the current session and draft a skill proposal. --- ## Step 1: Gather Session Context Read these sources to understand what's been happening: 1. Today's `.multiplai/diary/` entry — session summaries and key actions 2. Recent `.multiplai/learnings/` files — patterns and observations from recent sessions 3. The current conversation context (already available) Focus on: Which tool sequences recur? Which bash commands repeat? Which file types get edited in the same order? What workflows span multiple steps? ### Optional: `--from-session` — extract from a real trajectory Diary entries and learnings are *summaries*. They record that a workflow happened, not the exact commands, flag values, and dead ends it went through — so a skill drafted from them reads plausibly and then fails on first use, because the details it needed were the ones summarization dropped. When the user says "make a skill out of what we just did", or passes `--from-session [<id-prefix>]`, read the raw transcript instead: ``` ls -t "$CLAUDE_CONFIG_DIR/projects/$(pwd | tr '/' '-')"/*.jsonl | head -5 ``` The newest file is the current session. Read the transcript and pull out the **actual trajectory**: every tool call in order, with its real arguments, an