# Consume Session > Pick up a conversation that another Claude Code profile / subscription published into the shared staging directory (~/claude-profiles/shared-conversation/), so this session inherits what the other one already worked out. Use whenever the user wants to load, import, absorb, continue, or resume a session that came from their other account — phrasings like "load the session I shared from my work profile", "pick up where claude-personal left off", "consume ", "import the shared conversation", "what did my other account already do on this", "read the handoff transcript", or just pasting a bare session UUID and asking you to carry on with it. Also trigger when the user says they published a session from another profile and now wants it here, or asks what shared sessions are waiting to be picked up. Offers two paths: a cheap markdown digest read into this session, or installing the raw transcript so `claude --resume ` replays it in full. Always ends by asking whether to delete or keep the staged file - Skill: `labspangaea/consume-session` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add labspangaea/consume-session` - Raw SKILL.md: https://api.skillmd.com/api/skills/labspangaea/consume-session/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Docs & Writing - Author: labspangaea (https://skillmd.com/u/labspangaea) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/labspangaea/consume-session --- # Consume a session published by another profile `PLUGIN_DIR` below is this skill's own directory — the `consume-session/` folder holding this file, whether that is inside the installed plugin or a standalone `~/.agents/skills/consume-session`. ## Step 1 — find out what is waiting ```bash python3 PLUGIN_DIR/scripts/consume_session.py list ``` Each row is a session id, its size, and its title or working directory. If the user already gave you an id, skip straight to step 2. If the list is empty, the other profile has not published yet — tell them to run `/claude-profiles:publish-session` over there first. ## Step 2 — pick digest or resume These cost very different amounts, so choose deliberately rather than defaulting to whichever is mentioned first. **Digest (the normal choice).** Flattens the transcript to the conversation itself — human turns, Claude's prose, one line per tool call — and drops thinking blocks, tool results and subagent chatter, which is where the overwhelming majority of a transcript's bytes live. A 2 MB transcript typically lands around 20 KB. It works immediately, in this session, with no restart. ```bash python3 PLUGIN_DIR/scripts/consume_session.py digest ``` Then read the file it names. If it is very long, read the tail first — the end of a session is where the conclusions are — and page backwards only if you need the reasoning behind them. **Resume (full fidelity).** Copies the raw transcript into this profile's `projects/` directory so Claude Code can replay it exactly. ```bash python3 PLUGIN_DIR/scripts/consume_session.py resume ``` This does **not** affect the running session — it prints a `cd` and a `claude --resume ` to run in a fresh terminal. Say that plainly, because "resume" sounds like it should take effect here. Worth flagging too: replaying the whole transcript re-reads every token of it, which on the profile the user switched to precisely because of rate limits is a real cost before their first question. Reach for it when they need the exact file-edit history back, not just the knowledge. ## Step 3 — summarise what you inherited Do not just announce that the digest was read. Tell the user, in a few lines, what the other session established: what it was trying to do, what it settled on, and what it left unfinished. That summary is the actual deliverable — it is what lets them ask their next question without re-explaining anything. ## Step 4 — ask about the staged file Use `AskUserQuestion` to ask whether to delete or keep the staged transcript. Ask once, after the content has been used, so the answer is informed rather than a guess. If the user already answered this in their own request — "clean up after yourself", "delete it once you're done", "keep it, I'll consume it from my third profile too" — act on what they said instead of asking again. Re-asking a question the user has already answered reads as not having listened. Say which way you went, so a misread is easy to correct. - **Delete** — the hand-off is done. It is a full copy of a conversation sitting in the user's home directory, so removing it once consumed is the tidy default. - **Keep** — they may consume it again from a third profile, or want it as a record. Deleting removes the staged copy and its digest, and nothing else: ```bash python3 PLUGIN_DIR/scripts/consume_session.py delete ``` The original transcript in the publishing profile is untouched either way — the script refuses to delete anything outside the shared directory.